From 8f4d6c93a7be64e8b675657022bcdae2c21f8852 Mon Sep 17 00:00:00 2001 From: pinto0309 Date: Wed, 8 Nov 2023 00:57:47 +0900 Subject: [PATCH] Gold-YOLO-Head --- 421_Gold-YOLO-Head/LICENSE | 673 ++++++++++++++++++ 421_Gold-YOLO-Head/README.md | 1 + 421_Gold-YOLO-Head/demo/demo_goldyolo_onnx.py | 350 +++++++++ .../demo/demo_goldyolo_onnx_image.py | 314 ++++++++ 421_Gold-YOLO-Head/download_l.sh | 7 + 421_Gold-YOLO-Head/download_m.sh | 7 + 421_Gold-YOLO-Head/download_n.sh | 7 + 421_Gold-YOLO-Head/download_s.sh | 7 + .../post_process_gen_tools/README.md | 209 ++++++ .../post_process_gen_tools/boxes_listup.py | 54 ++ .../post_process_gen_tools/convert_script.sh | 484 +++++++++++++ .../demo_goldyolo_onnx.py | 340 +++++++++ .../make_batch_initialize.py | 209 ++++++ .../make_box_gather_nd.py | 62 ++ .../make_boxes_scores.py | 97 +++ .../make_cxcywh_x1y1x2y2.py | 72 ++ .../make_cxcywh_y1x1y2x2.py | 78 ++ ...ch_nums_final_class_nums_final_box_nums.py | 61 ++ .../make_input_output_shape_update.py | 76 ++ .../make_nms_outputs_merge.py | 64 ++ .../make_score_gather_nd.py | 72 ++ 421_Gold-YOLO-Head/url.txt | 5 + README.md | 1 + 23 files changed, 3250 insertions(+) create mode 100644 421_Gold-YOLO-Head/LICENSE create mode 100644 421_Gold-YOLO-Head/README.md create mode 100644 421_Gold-YOLO-Head/demo/demo_goldyolo_onnx.py create mode 100644 421_Gold-YOLO-Head/demo/demo_goldyolo_onnx_image.py create mode 100755 421_Gold-YOLO-Head/download_l.sh create mode 100755 421_Gold-YOLO-Head/download_m.sh create mode 100755 421_Gold-YOLO-Head/download_n.sh create mode 100755 421_Gold-YOLO-Head/download_s.sh create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/README.md create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/boxes_listup.py create mode 100755 421_Gold-YOLO-Head/post_process_gen_tools/convert_script.sh create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/demo_goldyolo_onnx.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_batch_initialize.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_box_gather_nd.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_boxes_scores.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_cxcywh_x1y1x2y2.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_cxcywh_y1x1y2x2.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_final_batch_nums_final_class_nums_final_box_nums.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_input_output_shape_update.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_nms_outputs_merge.py create mode 100644 421_Gold-YOLO-Head/post_process_gen_tools/make_score_gather_nd.py create mode 100644 421_Gold-YOLO-Head/url.txt diff --git a/421_Gold-YOLO-Head/LICENSE b/421_Gold-YOLO-Head/LICENSE new file mode 100644 index 0000000000..3ab9d83ecc --- /dev/null +++ b/421_Gold-YOLO-Head/LICENSE @@ -0,0 +1,673 @@ + GNU GENERAL PUBLIC LICENSE + Version 3, 29 June 2007 + + Copyright (C) 2007 Free Software Foundation, Inc. + Everyone is permitted to copy and distribute verbatim copies + of this license document, but changing it is not allowed. + + Preamble + + The GNU General Public License is a free, copyleft license for +software and other kinds of works. + + The licenses for most software and other practical works are designed +to take away your freedom to share and change the works. By contrast, +the GNU General Public License is intended to guarantee your freedom to +share and change all versions of a program--to make sure it remains free +software for all its users. We, the Free Software Foundation, use the +GNU General Public License for most of our software; it applies also to +any other work released this way by its authors. You can apply it to +your programs, too. + + When we speak of free software, we are referring to freedom, not +price. Our General Public Licenses are designed to make sure that you +have the freedom to distribute copies of free software (and charge for +them if you wish), that you receive source code or can get it if you +want it, that you can change the software or use pieces of it in new +free programs, and that you know you can do these things. + + To protect your rights, we need to prevent others from denying you +these rights or asking you to surrender the rights. Therefore, you have +certain responsibilities if you distribute copies of the software, or if +you modify it: responsibilities to respect the freedom of others. + + For example, if you distribute copies of such a program, whether +gratis or for a fee, you must pass on to the recipients the same +freedoms that you received. You must make sure that they, too, receive +or can get the source code. And you must show them these terms so they +know their rights. + + Developers that use the GNU GPL protect your rights with two steps: +(1) assert copyright on the software, and (2) offer you this License +giving you legal permission to copy, distribute and/or modify it. + + For the developers' and authors' protection, the GPL clearly explains +that there is no warranty for this free software. For both users' and +authors' sake, the GPL requires that modified versions be marked as +changed, so that their problems will not be attributed erroneously to +authors of previous versions. + + Some devices are designed to deny users access to install or run +modified versions of the software inside them, although the manufacturer +can do so. This is fundamentally incompatible with the aim of +protecting users' freedom to change the software. The systematic +pattern of such abuse occurs in the area of products for individuals to +use, which is precisely where it is most unacceptable. Therefore, we +have designed this version of the GPL to prohibit the practice for those +products. If such problems arise substantially in other domains, we +stand ready to extend this provision to those domains in future versions +of the GPL, as needed to protect the freedom of users. + + Finally, every program is threatened constantly by software patents. +States should not allow patents to restrict development and use of +software on general-purpose computers, but in those that do, we wish to +avoid the special danger that patents applied to a free program could +make it effectively proprietary. To prevent this, the GPL assures that +patents cannot be used to render the program non-free. + + The precise terms and conditions for copying, distribution and +modification follow. + + TERMS AND CONDITIONS + + 0. Definitions. + + "This License" refers to version 3 of the GNU General Public License. + + "Copyright" also means copyright-like laws that apply to other kinds of +works, such as semiconductor masks. + + "The Program" refers to any copyrightable work licensed under this +License. Each licensee is addressed as "you". "Licensees" and +"recipients" may be individuals or organizations. + + To "modify" a work means to copy from or adapt all or part of the work +in a fashion requiring copyright permission, other than the making of an +exact copy. The resulting work is called a "modified version" of the +earlier work or a work "based on" the earlier work. + + A "covered work" means either the unmodified Program or a work based +on the Program. + + To "propagate" a work means to do anything with it that, without +permission, would make you directly or secondarily liable for +infringement under applicable copyright law, except executing it on a +computer or modifying a private copy. Propagation includes copying, +distribution (with or without modification), making available to the +public, and in some countries other activities as well. + + To "convey" a work means any kind of propagation that enables other +parties to make or receive copies. Mere interaction with a user through +a computer network, with no transfer of a copy, is not conveying. + + An interactive user interface displays "Appropriate Legal Notices" +to the extent that it includes a convenient and prominently visible +feature that (1) displays an appropriate copyright notice, and (2) +tells the user that there is no warranty for the work (except to the +extent that warranties are provided), that licensees may convey the +work under this License, and how to view a copy of this License. If +the interface presents a list of user commands or options, such as a +menu, a prominent item in the list meets this criterion. + + 1. Source Code. + + The "source code" for a work means the preferred form of the work +for making modifications to it. "Object code" means any non-source +form of a work. + + A "Standard Interface" means an interface that either is an official +standard defined by a recognized standards body, or, in the case of +interfaces specified for a particular programming language, one that +is widely used among developers working in that language. + + The "System Libraries" of an executable work include anything, other +than the work as a whole, that (a) is included in the normal form of +packaging a Major Component, but which is not part of that Major +Component, and (b) serves only to enable use of the work with that +Major Component, or to implement a Standard Interface for which an +implementation is available to the public in source code form. A +"Major Component", in this context, means a major essential component +(kernel, window system, and so on) of the specific operating system +(if any) on which the executable work runs, or a compiler used to +produce the work, or an object code interpreter used to run it. + + The "Corresponding Source" for a work in object code form means all +the source code needed to generate, install, and (for an executable +work) run the object code and to modify the work, including scripts to +control those activities. However, it does not include the work's +System Libraries, or general-purpose tools or generally available free +programs which are used unmodified in performing those activities but +which are not part of the work. For example, Corresponding Source +includes interface definition files associated with source files for +the work, and the source code for shared libraries and dynamically +linked subprograms that the work is specifically designed to require, +such as by intimate data communication or control flow between those +subprograms and other parts of the work. + + The Corresponding Source need not include anything that users +can regenerate automatically from other parts of the Corresponding +Source. + + The Corresponding Source for a work in source code form is that +same work. + + 2. Basic Permissions. + + All rights granted under this License are granted for the term of +copyright on the Program, and are irrevocable provided the stated +conditions are met. This License explicitly affirms your unlimited +permission to run the unmodified Program. The output from running a +covered work is covered by this License only if the output, given its +content, constitutes a covered work. This License acknowledges your +rights of fair use or other equivalent, as provided by copyright law. + + You may make, run and propagate covered works that you do not +convey, without conditions so long as your license otherwise remains +in force. You may convey covered works to others for the sole purpose +of having them make modifications exclusively for you, or provide you +with facilities for running those works, provided that you comply with +the terms of this License in conveying all material for which you do +not control copyright. Those thus making or running the covered works +for you must do so exclusively on your behalf, under your direction +and control, on terms that prohibit them from making any copies of +your copyrighted material outside their relationship with you. + + Conveying under any other circumstances is permitted solely under +the conditions stated below. Sublicensing is not allowed; section 10 +makes it unnecessary. + + 3. Protecting Users' Legal Rights From Anti-Circumvention Law. + + No covered work shall be deemed part of an effective technological +measure under any applicable law fulfilling obligations under article +11 of the WIPO copyright treaty adopted on 20 December 1996, or +similar laws prohibiting or restricting circumvention of such +measures. + + When you convey a covered work, you waive any legal power to forbid +circumvention of technological measures to the extent such circumvention +is effected by exercising rights under this License with respect to +the covered work, and you disclaim any intention to limit operation or +modification of the work as a means of enforcing, against the work's +users, your or third parties' legal rights to forbid circumvention of +technological measures. + + 4. Conveying Verbatim Copies. + + You may convey verbatim copies of the Program's source code as you +receive it, in any medium, provided that you conspicuously and +appropriately publish on each copy an appropriate copyright notice; +keep intact all notices stating that this License and any +non-permissive terms added in accord with section 7 apply to the code; +keep intact all notices of the absence of any warranty; and give all +recipients a copy of this License along with the Program. + + You may charge any price or no price for each copy that you convey, +and you may offer support or warranty protection for a fee. + + 5. Conveying Modified Source Versions. + + You may convey a work based on the Program, or the modifications to +produce it from the Program, in the form of source code under the +terms of section 4, provided that you also meet all of these conditions: + + a) The work must carry prominent notices stating that you modified + it, and giving a relevant date. + + b) The work must carry prominent notices stating that it is + released under this License and any conditions added under section + 7. This requirement modifies the requirement in section 4 to + "keep intact all notices". + + c) You must license the entire work, as a whole, under this + License to anyone who comes into possession of a copy. This + License will therefore apply, along with any applicable section 7 + additional terms, to the whole of the work, and all its parts, + regardless of how they are packaged. This License gives no + permission to license the work in any other way, but it does not + invalidate such permission if you have separately received it. + + d) If the work has interactive user interfaces, each must display + Appropriate Legal Notices; however, if the Program has interactive + interfaces that do not display Appropriate Legal Notices, your + work need not make them do so. + + A compilation of a covered work with other separate and independent +works, which are not by their nature extensions of the covered work, +and which are not combined with it such as to form a larger program, +in or on a volume of a storage or distribution medium, is called an +"aggregate" if the compilation and its resulting copyright are not +used to limit the access or legal rights of the compilation's users +beyond what the individual works permit. Inclusion of a covered work +in an aggregate does not cause this License to apply to the other +parts of the aggregate. + + 6. Conveying Non-Source Forms. + + You may convey a covered work in object code form under the terms +of sections 4 and 5, provided that you also convey the +machine-readable Corresponding Source under the terms of this License, +in one of these ways: + + a) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by the + Corresponding Source fixed on a durable physical medium + customarily used for software interchange. + + b) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by a + written offer, valid for at least three years and valid for as + long as you offer spare parts or customer support for that product + model, to give anyone who possesses the object code either (1) a + copy of the Corresponding Source for all the software in the + product that is covered by this License, on a durable physical + medium customarily used for software interchange, for a price no + more than your reasonable cost of physically performing this + conveying of source, or (2) access to copy the + Corresponding Source from a network server at no charge. + + c) Convey individual copies of the object code with a copy of the + written offer to provide the Corresponding Source. This + alternative is allowed only occasionally and noncommercially, and + only if you received the object code with such an offer, in accord + with subsection 6b. + + d) Convey the object code by offering access from a designated + place (gratis or for a charge), and offer equivalent access to the + Corresponding Source in the same way through the same place at no + further charge. You need not require recipients to copy the + Corresponding Source along with the object code. If the place to + copy the object code is a network server, the Corresponding Source + may be on a different server (operated by you or a third party) + that supports equivalent copying facilities, provided you maintain + clear directions next to the object code saying where to find the + Corresponding Source. Regardless of what server hosts the + Corresponding Source, you remain obligated to ensure that it is + available for as long as needed to satisfy these requirements. + + e) Convey the object code using peer-to-peer transmission, provided + you inform other peers where the object code and Corresponding + Source of the work are being offered to the general public at no + charge under subsection 6d. + + A separable portion of the object code, whose source code is excluded +from the Corresponding Source as a System Library, need not be +included in conveying the object code work. + + A "User Product" is either (1) a "consumer product", which means any +tangible personal property which is normally used for personal, family, +or household purposes, or (2) anything designed or sold for incorporation +into a dwelling. In determining whether a product is a consumer product, +doubtful cases shall be resolved in favor of coverage. For a particular +product received by a particular user, "normally used" refers to a +typical or common use of that class of product, regardless of the status +of the particular user or of the way in which the particular user +actually uses, or expects or is expected to use, the product. A product +is a consumer product regardless of whether the product has substantial +commercial, industrial or non-consumer uses, unless such uses represent +the only significant mode of use of the product. + + "Installation Information" for a User Product means any methods, +procedures, authorization keys, or other information required to install +and execute modified versions of a covered work in that User Product from +a modified version of its Corresponding Source. The information must +suffice to ensure that the continued functioning of the modified object +code is in no case prevented or interfered with solely because +modification has been made. + + If you convey an object code work under this section in, or with, or +specifically for use in, a User Product, and the conveying occurs as +part of a transaction in which the right of possession and use of the +User Product is transferred to the recipient in perpetuity or for a +fixed term (regardless of how the transaction is characterized), the +Corresponding Source conveyed under this section must be accompanied +by the Installation Information. But this requirement does not apply +if neither you nor any third party retains the ability to install +modified object code on the User Product (for example, the work has +been installed in ROM). + + The requirement to provide Installation Information does not include a +requirement to continue to provide support service, warranty, or updates +for a work that has been modified or installed by the recipient, or for +the User Product in which it has been modified or installed. Access to a +network may be denied when the modification itself materially and +adversely affects the operation of the network or violates the rules and +protocols for communication across the network. + + Corresponding Source conveyed, and Installation Information provided, +in accord with this section must be in a format that is publicly +documented (and with an implementation available to the public in +source code form), and must require no special password or key for +unpacking, reading or copying. + + 7. Additional Terms. + + "Additional permissions" are terms that supplement the terms of this +License by making exceptions from one or more of its conditions. +Additional permissions that are applicable to the entire Program shall +be treated as though they were included in this License, to the extent +that they are valid under applicable law. If additional permissions +apply only to part of the Program, that part may be used separately +under those permissions, but the entire Program remains governed by +this License without regard to the additional permissions. + + When you convey a copy of a covered work, you may at your option +remove any additional permissions from that copy, or from any part of +it. (Additional permissions may be written to require their own +removal in certain cases when you modify the work.) You may place +additional permissions on material, added by you to a covered work, +for which you have or can give appropriate copyright permission. + + Notwithstanding any other provision of this License, for material you +add to a covered work, you may (if authorized by the copyright holders of +that material) supplement the terms of this License with terms: + + a) Disclaiming warranty or limiting liability differently from the + terms of sections 15 and 16 of this License; or + + b) Requiring preservation of specified reasonable legal notices or + author attributions in that material or in the Appropriate Legal + Notices displayed by works containing it; or + + c) Prohibiting misrepresentation of the origin of that material, or + requiring that modified versions of such material be marked in + reasonable ways as different from the original version; or + + d) Limiting the use for publicity purposes of names of licensors or + authors of the material; or + + e) Declining to grant rights under trademark law for use of some + trade names, trademarks, or service marks; or + + f) Requiring indemnification of licensors and authors of that + material by anyone who conveys the material (or modified versions of + it) with contractual assumptions of liability to the recipient, for + any liability that these contractual assumptions directly impose on + those licensors and authors. + + All other non-permissive additional terms are considered "further +restrictions" within the meaning of section 10. If the Program as you +received it, or any part of it, contains a notice stating that it is +governed by this License along with a term that is a further +restriction, you may remove that term. If a license document contains +a further restriction but permits relicensing or conveying under this +License, you may add to a covered work material governed by the terms +of that license document, provided that the further restriction does +not survive such relicensing or conveying. + + If you add terms to a covered work in accord with this section, you +must place, in the relevant source files, a statement of the +additional terms that apply to those files, or a notice indicating +where to find the applicable terms. + + Additional terms, permissive or non-permissive, may be stated in the +form of a separately written license, or stated as exceptions; +the above requirements apply either way. + + 8. Termination. + + You may not propagate or modify a covered work except as expressly +provided under this License. Any attempt otherwise to propagate or +modify it is void, and will automatically terminate your rights under +this License (including any patent licenses granted under the third +paragraph of section 11). + + However, if you cease all violation of this License, then your +license from a particular copyright holder is reinstated (a) +provisionally, unless and until the copyright holder explicitly and +finally terminates your license, and (b) permanently, if the copyright +holder fails to notify you of the violation by some reasonable means +prior to 60 days after the cessation. + + Moreover, your license from a particular copyright holder is +reinstated permanently if the copyright holder notifies you of the +violation by some reasonable means, this is the first time you have +received notice of violation of this License (for any work) from that +copyright holder, and you cure the violation prior to 30 days after +your receipt of the notice. + + Termination of your rights under this section does not terminate the +licenses of parties who have received copies or rights from you under +this License. If your rights have been terminated and not permanently +reinstated, you do not qualify to receive new licenses for the same +material under section 10. + + 9. Acceptance Not Required for Having Copies. + + You are not required to accept this License in order to receive or +run a copy of the Program. Ancillary propagation of a covered work +occurring solely as a consequence of using peer-to-peer transmission +to receive a copy likewise does not require acceptance. However, +nothing other than this License grants you permission to propagate or +modify any covered work. These actions infringe copyright if you do +not accept this License. Therefore, by modifying or propagating a +covered work, you indicate your acceptance of this License to do so. + + 10. Automatic Licensing of Downstream Recipients. + + Each time you convey a covered work, the recipient automatically +receives a license from the original licensors, to run, modify and +propagate that work, subject to this License. You are not responsible +for enforcing compliance by third parties with this License. + + An "entity transaction" is a transaction transferring control of an +organization, or substantially all assets of one, or subdividing an +organization, or merging organizations. If propagation of a covered +work results from an entity transaction, each party to that +transaction who receives a copy of the work also receives whatever +licenses to the work the party's predecessor in interest had or could +give under the previous paragraph, plus a right to possession of the +Corresponding Source of the work from the predecessor in interest, if +the predecessor has it or can get it with reasonable efforts. + + You may not impose any further restrictions on the exercise of the +rights granted or affirmed under this License. For example, you may +not impose a license fee, royalty, or other charge for exercise of +rights granted under this License, and you may not initiate litigation +(including a cross-claim or counterclaim in a lawsuit) alleging that +any patent claim is infringed by making, using, selling, offering for +sale, or importing the Program or any portion of it. + + 11. Patents. + + A "contributor" is a copyright holder who authorizes use under this +License of the Program or a work on which the Program is based. The +work thus licensed is called the contributor's "contributor version". + + A contributor's "essential patent claims" are all patent claims +owned or controlled by the contributor, whether already acquired or +hereafter acquired, that would be infringed by some manner, permitted +by this License, of making, using, or selling its contributor version, +but do not include claims that would be infringed only as a +consequence of further modification of the contributor version. For +purposes of this definition, "control" includes the right to grant +patent sublicenses in a manner consistent with the requirements of +this License. + + Each contributor grants you a non-exclusive, worldwide, royalty-free +patent license under the contributor's essential patent claims, to +make, use, sell, offer for sale, import and otherwise run, modify and +propagate the contents of its contributor version. + + In the following three paragraphs, a "patent license" is any express +agreement or commitment, however denominated, not to enforce a patent +(such as an express permission to practice a patent or covenant not to +sue for patent infringement). To "grant" such a patent license to a +party means to make such an agreement or commitment not to enforce a +patent against the party. + + If you convey a covered work, knowingly relying on a patent license, +and the Corresponding Source of the work is not available for anyone +to copy, free of charge and under the terms of this License, through a +publicly available network server or other readily accessible means, +then you must either (1) cause the Corresponding Source to be so +available, or (2) arrange to deprive yourself of the benefit of the +patent license for this particular work, or (3) arrange, in a manner +consistent with the requirements of this License, to extend the patent +license to downstream recipients. "Knowingly relying" means you have +actual knowledge that, but for the patent license, your conveying the +covered work in a country, or your recipient's use of the covered work +in a country, would infringe one or more identifiable patents in that +country that you have reason to believe are valid. + + If, pursuant to or in connection with a single transaction or +arrangement, you convey, or propagate by procuring conveyance of, a +covered work, and grant a patent license to some of the parties +receiving the covered work authorizing them to use, propagate, modify +or convey a specific copy of the covered work, then the patent license +you grant is automatically extended to all recipients of the covered +work and works based on it. + + A patent license is "discriminatory" if it does not include within +the scope of its coverage, prohibits the exercise of, or is +conditioned on the non-exercise of one or more of the rights that are +specifically granted under this License. You may not convey a covered +work if you are a party to an arrangement with a third party that is +in the business of distributing software, under which you make payment +to the third party based on the extent of your activity of conveying +the work, and under which the third party grants, to any of the +parties who would receive the covered work from you, a discriminatory +patent license (a) in connection with copies of the covered work +conveyed by you (or copies made from those copies), or (b) primarily +for and in connection with specific products or compilations that +contain the covered work, unless you entered into that arrangement, +or that patent license was granted, prior to 28 March 2007. + + Nothing in this License shall be construed as excluding or limiting +any implied license or other defenses to infringement that may +otherwise be available to you under applicable patent law. + + 12. No Surrender of Others' Freedom. + + If conditions are imposed on you (whether by court order, agreement or +otherwise) that contradict the conditions of this License, they do not +excuse you from the conditions of this License. If you cannot convey a +covered work so as to satisfy simultaneously your obligations under this +License and any other pertinent obligations, then as a consequence you may +not convey it at all. For example, if you agree to terms that obligate you +to collect a royalty for further conveying from those to whom you convey +the Program, the only way you could satisfy both those terms and this +License would be to refrain entirely from conveying the Program. + + 13. Use with the GNU Affero General Public License. + + Notwithstanding any other provision of this License, you have +permission to link or combine any covered work with a work licensed +under version 3 of the GNU Affero General Public License into a single +combined work, and to convey the resulting work. The terms of this +License will continue to apply to the part which is the covered work, +but the special requirements of the GNU Affero General Public License, +section 13, concerning interaction through a network will apply to the +combination as such. + + 14. Revised Versions of this License. + + The Free Software Foundation may publish revised and/or new versions of +the GNU General Public License from time to time. Such new versions will +be similar in spirit to the present version, but may differ in detail to +address new problems or concerns. + + Each version is given a distinguishing version number. If the +Program specifies that a certain numbered version of the GNU General +Public License "or any later version" applies to it, you have the +option of following the terms and conditions either of that numbered +version or of any later version published by the Free Software +Foundation. If the Program does not specify a version number of the +GNU General Public License, you may choose any version ever published +by the Free Software Foundation. + + If the Program specifies that a proxy can decide which future +versions of the GNU General Public License can be used, that proxy's +public statement of acceptance of a version permanently authorizes you +to choose that version for the Program. + + Later license versions may give you additional or different +permissions. However, no additional obligations are imposed on any +author or copyright holder as a result of your choosing to follow a +later version. + + 15. Disclaimer of Warranty. + + THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY +APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT +HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY +OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, +THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM +IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF +ALL NECESSARY SERVICING, REPAIR OR CORRECTION. + + 16. Limitation of Liability. + + IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING +WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS +THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY +GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE +USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF +DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD +PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), +EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF +SUCH DAMAGES. + + 17. Interpretation of Sections 15 and 16. + + If the disclaimer of warranty and limitation of liability provided +above cannot be given local legal effect according to their terms, +reviewing courts shall apply local law that most closely approximates +an absolute waiver of all civil liability in connection with the +Program, unless a warranty or assumption of liability accompanies a +copy of the Program in return for a fee. + + END OF TERMS AND CONDITIONS + + How to Apply These Terms to Your New Programs + + If you develop a new program, and you want it to be of the greatest +possible use to the public, the best way to achieve this is to make it +free software which everyone can redistribute and change under these terms. + + To do so, attach the following notices to the program. It is safest +to attach them to the start of each source file to most effectively +state the exclusion of warranty; and each file should have at least +the "copyright" line and a pointer to where the full notice is found. + + + Copyright (C) + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + +Also add information on how to contact you by electronic and paper mail. + + If the program does terminal interaction, make it output a short +notice like this when it starts in an interactive mode: + + Copyright (C) + This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. + This is free software, and you are welcome to redistribute it + under certain conditions; type `show c' for details. + +The hypothetical commands `show w' and `show c' should show the appropriate +parts of the General Public License. Of course, your program's commands +might be different; for a GUI interface, you would use an "about box". + + You should also get your employer (if you work as a programmer) or school, +if any, to sign a "copyright disclaimer" for the program, if necessary. +For more information on this, and how to apply and follow the GNU GPL, see +. + + The GNU General Public License does not permit incorporating your program +into proprietary programs. If your program is a subroutine library, you +may consider it more useful to permit linking proprietary applications with +the library. If this is what you want to do, use the GNU Lesser General +Public License instead of this License. But first, please read \ No newline at end of file diff --git a/421_Gold-YOLO-Head/README.md b/421_Gold-YOLO-Head/README.md new file mode 100644 index 0000000000..0e30879d1d --- /dev/null +++ b/421_Gold-YOLO-Head/README.md @@ -0,0 +1 @@ +# Note diff --git a/421_Gold-YOLO-Head/demo/demo_goldyolo_onnx.py b/421_Gold-YOLO-Head/demo/demo_goldyolo_onnx.py new file mode 100644 index 0000000000..acbc057a48 --- /dev/null +++ b/421_Gold-YOLO-Head/demo/demo_goldyolo_onnx.py @@ -0,0 +1,350 @@ +#!/usr/bin/env python + +import copy +import cv2 +import time +import numpy as np +import onnxruntime +from argparse import ArgumentParser +from typing import Tuple, Optional, List + + +class GoldYOLOONNX(object): + def __init__( + self, + model_path: Optional[str] = 'gold_yolo_n_hand_post_0333_0.4040_1x3x480x640.onnx', + class_score_th: Optional[float] = 0.35, + providers: Optional[List] = [ + # ( + # 'TensorrtExecutionProvider', { + # 'trt_engine_cache_enable': True, + # 'trt_engine_cache_path': '.', + # 'trt_fp16_enable': True, + # } + # ), + 'CUDAExecutionProvider', + 'CPUExecutionProvider', + ], + ): + """GoldYOLOONNX + + Parameters + ---------- + model_path: Optional[str] + ONNX file path for GoldYOLO + + class_score_th: Optional[float] + Score threshold. Default: 0.35 + + providers: Optional[List] + Name of onnx execution providers + Default: + [ + ( + 'TensorrtExecutionProvider', { + 'trt_engine_cache_enable': True, + 'trt_engine_cache_path': '.', + 'trt_fp16_enable': True, + } + ), + 'CUDAExecutionProvider', + 'CPUExecutionProvider', + ] + """ + # Threshold + self.class_score_th = class_score_th + + # Model loading + session_option = onnxruntime.SessionOptions() + session_option.log_severity_level = 3 + self.onnx_session = onnxruntime.InferenceSession( + model_path, + sess_options=session_option, + providers=providers, + ) + self.providers = self.onnx_session.get_providers() + + self.input_shapes = [ + input.shape for input in self.onnx_session.get_inputs() + ] + self.input_names = [ + input.name for input in self.onnx_session.get_inputs() + ] + self.output_names = [ + output.name for output in self.onnx_session.get_outputs() + ] + + + def __call__( + self, + image: np.ndarray, + ) -> Tuple[np.ndarray, np.ndarray]: + """YOLOv7ONNX + + Parameters + ---------- + image: np.ndarray + Entire image + + Returns + ------- + boxes: np.ndarray + Predicted boxes: [N, y1, x1, y2, x2] + + scores: np.ndarray + Predicted box scores: [N, score] + """ + temp_image = copy.deepcopy(image) + + # PreProcess + resized_image = self.__preprocess( + temp_image, + ) + + # Inference + inferece_image = np.asarray([resized_image], dtype=np.float32) + boxes = self.onnx_session.run( + self.output_names, + {input_name: inferece_image for input_name in self.input_names}, + )[0] + + # PostProcess + result_boxes, result_scores = \ + self.__postprocess( + image=temp_image, + boxes=boxes, + ) + + return result_boxes, result_scores + + + def __preprocess( + self, + image: np.ndarray, + swap: Optional[Tuple[int,int,int]] = (2, 0, 1), + ) -> np.ndarray: + """__preprocess + + Parameters + ---------- + image: np.ndarray + Entire image + + swap: tuple + HWC to CHW: (2,0,1) + CHW to HWC: (1,2,0) + HWC to HWC: (0,1,2) + CHW to CHW: (0,1,2) + + Returns + ------- + resized_image: np.ndarray + Resized and normalized image. + """ + # Normalization + BGR->RGB + resized_image = cv2.resize( + image, + ( + int(self.input_shapes[0][3]), + int(self.input_shapes[0][2]), + ) + ) + resized_image = np.divide(resized_image, 255.0) + resized_image = resized_image[..., ::-1] + resized_image = resized_image.transpose(swap) + resized_image = np.ascontiguousarray( + resized_image, + dtype=np.float32, + ) + return resized_image + + + def __postprocess( + self, + image: np.ndarray, + boxes: np.ndarray, + ) -> Tuple[np.ndarray, np.ndarray]: + """__postprocess + + Parameters + ---------- + image: np.ndarray + Entire image. + + boxes: np.ndarray + float32[N, 7] + + Returns + ------- + result_boxes: np.ndarray + Predicted boxes: [N, y1, x1, y2, x2] + + result_scores: np.ndarray + Predicted box confs: [N, score] + """ + image_height = image.shape[0] + image_width = image.shape[1] + + """ + Detector is + N -> Number of boxes detected + batchno -> always 0: BatchNo.0 + + batchno_classid_y1x1y2x2_score: float32[N,7] + """ + result_boxes = [] + result_scores = [] + if len(boxes) > 0: + scores = boxes[:, 6:7] + keep_idxs = scores[:, 0] > self.class_score_th + scores_keep = scores[keep_idxs, :] + boxes_keep = boxes[keep_idxs, :] + + if len(boxes_keep) > 0: + for box, score in zip(boxes_keep, scores_keep): + x_min = int(max(box[2], 0) * image_width / self.input_shapes[0][3]) + y_min = int(max(box[3], 0) * image_height / self.input_shapes[0][2]) + x_max = int(min(box[4], self.input_shapes[0][3]) * image_width / self.input_shapes[0][3]) + y_max = int(min(box[5], self.input_shapes[0][2]) * image_height / self.input_shapes[0][2]) + + result_boxes.append( + [x_min, y_min, x_max, y_max] + ) + result_scores.append( + score + ) + + return np.asarray(result_boxes), np.asarray(result_scores) + + +def is_parsable_to_int(s): + try: + int(s) + return True + except ValueError: + return False + + +def main(): + parser = ArgumentParser() + parser.add_argument( + '-m', + '--model', + type=str, + default='gold_yolo_n_hand_post_0333_0.4040_1x3x480x640.onnx', + ) + parser.add_argument( + '-v', + '--video', + type=str, + default="0", + ) + args = parser.parse_args() + + model = GoldYOLOONNX( + model_path=args.model, + ) + + cap = cv2.VideoCapture( + int(args.video) if is_parsable_to_int(args.video) else args.video + ) + cap_fps = cap.get(cv2.CAP_PROP_FPS) + w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) + h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) + fourcc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v') + video_writer = cv2.VideoWriter( + filename='output.mp4', + fourcc=fourcc, + fps=cap_fps, + frameSize=(w, h), + ) + + while cap.isOpened(): + res, image = cap.read() + if not res: + break + + debug_image = copy.deepcopy(image) + + start_time = time.time() + boxes, scores = model(debug_image) + elapsed_time = time.time() - start_time + fps = 1 / elapsed_time + cv2.putText( + debug_image, + f'{fps:.1f} FPS (inferece + post-process)', + (10, 30), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (255, 255, 255), + 2, + cv2.LINE_AA, + ) + cv2.putText( + debug_image, + f'{fps:.1f} FPS (inferece + post-process)', + (10, 30), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (0, 0, 255), + 1, + cv2.LINE_AA, + ) + + for box, score in zip(boxes, scores): + cv2.rectangle( + debug_image, + (box[0], box[1]), + (box[2], box[3]), + (255,255,255), + 2, + ) + cv2.rectangle( + debug_image, + (box[0], box[1]), + (box[2], box[3]), + (0,0,255), + 1, + ) + cv2.putText( + debug_image, + f'{score[0]:.2f}', + ( + box[0], + box[1]-10 if box[1]-10 > 0 else 10 + ), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (255, 255, 255), + 2, + cv2.LINE_AA, + ) + cv2.putText( + debug_image, + f'{score[0]:.2f}', + ( + box[0], + box[1]-10 if box[1]-10 > 0 else 10 + ), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (0, 0, 255), + 1, + cv2.LINE_AA, + ) + + key = cv2.waitKey(1) + if key == 27: # ESC + break + + cv2.imshow("test", debug_image) + video_writer.write(debug_image) + + if video_writer: + video_writer.release() + + if cap: + cap.release() + +if __name__ == "__main__": + main() diff --git a/421_Gold-YOLO-Head/demo/demo_goldyolo_onnx_image.py b/421_Gold-YOLO-Head/demo/demo_goldyolo_onnx_image.py new file mode 100644 index 0000000000..7536d66fba --- /dev/null +++ b/421_Gold-YOLO-Head/demo/demo_goldyolo_onnx_image.py @@ -0,0 +1,314 @@ +#!/usr/bin/env python + +import os +import copy +import cv2 +from tqdm import tqdm +import glob +import numpy as np +import onnxruntime +from argparse import ArgumentParser +from typing import Tuple, Optional, List + + +class GoldYOLOONNX(object): + def __init__( + self, + model_path: Optional[str] = 'gold_yolo_n_hand_post_0333_0.4040_1x3x480x640.onnx', + class_score_th: Optional[float] = 0.35, + providers: Optional[List] = [ + ( + 'TensorrtExecutionProvider', { + 'trt_engine_cache_enable': True, + 'trt_engine_cache_path': '.', + 'trt_fp16_enable': True, + } + ), + 'CUDAExecutionProvider', + 'CPUExecutionProvider', + ], + ): + """GoldYOLOONNX + + Parameters + ---------- + model_path: Optional[str] + ONNX file path for GoldYOLO + + class_score_th: Optional[float] + Score threshold. Default: 0.35 + + providers: Optional[List] + Name of onnx execution providers + Default: + [ + ( + 'TensorrtExecutionProvider', { + 'trt_engine_cache_enable': True, + 'trt_engine_cache_path': '.', + 'trt_fp16_enable': True, + } + ), + 'CUDAExecutionProvider', + 'CPUExecutionProvider', + ] + """ + # Threshold + self.class_score_th = class_score_th + + # Model loading + session_option = onnxruntime.SessionOptions() + session_option.log_severity_level = 3 + self.onnx_session = onnxruntime.InferenceSession( + model_path, + sess_options=session_option, + providers=providers, + ) + self.providers = self.onnx_session.get_providers() + + self.input_shapes = [ + input.shape for input in self.onnx_session.get_inputs() + ] + self.input_names = [ + input.name for input in self.onnx_session.get_inputs() + ] + self.output_names = [ + output.name for output in self.onnx_session.get_outputs() + ] + + + def __call__( + self, + image: np.ndarray, + ) -> Tuple[np.ndarray, np.ndarray]: + """YOLOv7ONNX + + Parameters + ---------- + image: np.ndarray + Entire image + + Returns + ------- + boxes: np.ndarray + Predicted boxes: [N, y1, x1, y2, x2] + + scores: np.ndarray + Predicted box scores: [N, score] + """ + temp_image = copy.deepcopy(image) + + # PreProcess + resized_image = self.__preprocess( + temp_image, + ) + + # Inference + inferece_image = np.asarray([resized_image], dtype=np.float32) + boxes = self.onnx_session.run( + self.output_names, + {input_name: inferece_image for input_name in self.input_names}, + )[0] + + # PostProcess + result_boxes, result_scores = \ + self.__postprocess( + image=temp_image, + boxes=boxes, + ) + + return result_boxes, result_scores + + + def __preprocess( + self, + image: np.ndarray, + swap: Optional[Tuple[int,int,int]] = (2, 0, 1), + ) -> np.ndarray: + """__preprocess + + Parameters + ---------- + image: np.ndarray + Entire image + + swap: tuple + HWC to CHW: (2,0,1) + CHW to HWC: (1,2,0) + HWC to HWC: (0,1,2) + CHW to CHW: (0,1,2) + + Returns + ------- + resized_image: np.ndarray + Resized and normalized image. + """ + # Normalization + BGR->RGB + resized_image = cv2.resize( + image, + ( + int(self.input_shapes[0][3]), + int(self.input_shapes[0][2]), + ) + ) + resized_image = np.divide(resized_image, 255.0) + resized_image = resized_image[..., ::-1] + resized_image = resized_image.transpose(swap) + resized_image = np.ascontiguousarray( + resized_image, + dtype=np.float32, + ) + return resized_image + + + def __postprocess( + self, + image: np.ndarray, + boxes: np.ndarray, + ) -> Tuple[np.ndarray, np.ndarray]: + """__postprocess + + Parameters + ---------- + image: np.ndarray + Entire image. + + boxes: np.ndarray + float32[N, 7] + + Returns + ------- + result_boxes: np.ndarray + Predicted boxes: [N, y1, x1, y2, x2] + + result_scores: np.ndarray + Predicted box confs: [N, score] + """ + image_height = image.shape[0] + image_width = image.shape[1] + + """ + Detector is + N -> Number of boxes detected + batchno -> always 0: BatchNo.0 + + batchno_classid_y1x1y2x2_score: float32[N,7] + """ + result_boxes = [] + result_scores = [] + if len(boxes) > 0: + scores = boxes[:, 6:7] + keep_idxs = scores[:, 0] > self.class_score_th + scores_keep = scores[keep_idxs, :] + boxes_keep = boxes[keep_idxs, :] + + if len(boxes_keep) > 0: + for box, score in zip(boxes_keep, scores_keep): + x_min = int(max(box[2], 0) * image_width / self.input_shapes[0][3]) + y_min = int(max(box[3], 0) * image_height / self.input_shapes[0][2]) + x_max = int(min(box[4], self.input_shapes[0][3]) * image_width / self.input_shapes[0][3]) + y_max = int(min(box[5], self.input_shapes[0][2]) * image_height / self.input_shapes[0][2]) + + result_boxes.append( + [x_min, y_min, x_max, y_max] + ) + result_scores.append( + score + ) + + return np.asarray(result_boxes), np.asarray(result_scores) + + +def is_parsable_to_int(s): + try: + int(s) + return True + except ValueError: + return False + + +def main(): + parser = ArgumentParser() + parser.add_argument( + '-m', + '--model', + type=str, + default='gold_yolo_n_hand_post_0333_0.4040_1x3x480x640.onnx', + ) + parser.add_argument( + '-i', + '--images_path', + type=str, + default="./00_COCO-Hand-S_base", + ) + parser.add_argument( + '-o', + '--output_path', + type=str, + default="./output", + ) + args = parser.parse_args() + + model = GoldYOLOONNX( + model_path=args.model, + ) + + files = sorted(glob.glob(f"{args.images_path}/*.jpg")) + os.makedirs(args.output_path, exist_ok=True) + + for file in tqdm(files, dynamic_ncols=True): + image = cv2.imread(file) + debug_image = copy.deepcopy(image) + boxes, scores = model(debug_image) + + for box, score in zip(boxes, scores): + cv2.rectangle( + debug_image, + (box[0], box[1]), + (box[2], box[3]), + (255,255,255), + 2, + ) + cv2.rectangle( + debug_image, + (box[0], box[1]), + (box[2], box[3]), + (0,0,255), + 1, + ) + cv2.putText( + debug_image, + f'{score[0]:.2f}', + ( + box[0], + box[1]-10 if box[1]-10 > 0 else 10 + ), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (255, 255, 255), + 2, + cv2.LINE_AA, + ) + cv2.putText( + debug_image, + f'{score[0]:.2f}', + ( + box[0], + box[1]-10 if box[1]-10 > 0 else 10 + ), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (0, 0, 255), + 1, + cv2.LINE_AA, + ) + cv2.imwrite(f'{args.output_path}/{os.path.basename(file)}', debug_image) + + key = cv2.waitKey(1) + if key == 27: # ESC + break + + cv2.imshow("test", debug_image) + +if __name__ == "__main__": + main() diff --git a/421_Gold-YOLO-Head/download_l.sh b/421_Gold-YOLO-Head/download_l.sh new file mode 100755 index 0000000000..1e35406ad5 --- /dev/null +++ b/421_Gold-YOLO-Head/download_l.sh @@ -0,0 +1,7 @@ +#!/bin/bash + +curl "https://s3.ap-northeast-2.wasabisys.com/pinto-model-zoo/421_Gold-YOLO-Head/resources_l.tar.gz" -o resources.tar.gz +tar -zxvf resources.tar.gz +rm resources.tar.gz + +echo Download finished. diff --git a/421_Gold-YOLO-Head/download_m.sh b/421_Gold-YOLO-Head/download_m.sh new file mode 100755 index 0000000000..295616a910 --- /dev/null +++ b/421_Gold-YOLO-Head/download_m.sh @@ -0,0 +1,7 @@ +#!/bin/bash + +curl "https://s3.ap-northeast-2.wasabisys.com/pinto-model-zoo/421_Gold-YOLO-Head/resources_m.tar.gz" -o resources.tar.gz +tar -zxvf resources.tar.gz +rm resources.tar.gz + +echo Download finished. diff --git a/421_Gold-YOLO-Head/download_n.sh b/421_Gold-YOLO-Head/download_n.sh new file mode 100755 index 0000000000..591c68db08 --- /dev/null +++ b/421_Gold-YOLO-Head/download_n.sh @@ -0,0 +1,7 @@ +#!/bin/bash + +curl "https://s3.ap-northeast-2.wasabisys.com/pinto-model-zoo/421_Gold-YOLO-Head/resources_n.tar.gz" -o resources.tar.gz +tar -zxvf resources.tar.gz +rm resources.tar.gz + +echo Download finished. diff --git a/421_Gold-YOLO-Head/download_s.sh b/421_Gold-YOLO-Head/download_s.sh new file mode 100755 index 0000000000..e8d6c5c473 --- /dev/null +++ b/421_Gold-YOLO-Head/download_s.sh @@ -0,0 +1,7 @@ +#!/bin/bash + +curl "https://s3.ap-northeast-2.wasabisys.com/pinto-model-zoo/421_Gold-YOLO-Head/resources_s.tar.gz" -o resources.tar.gz +tar -zxvf resources.tar.gz +rm resources.tar.gz + +echo Download finished. diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/README.md b/421_Gold-YOLO-Head/post_process_gen_tools/README.md new file mode 100644 index 0000000000..6cf00fd90a --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/README.md @@ -0,0 +1,209 @@ +# Note +- INPUTS + - `predictions`: `float32 [batches, boxes, 5 + classes]` + + * 5 = [center_x, center_y, width, height, score] +- OUTPUTS + - `batchno_classid_x1y1x2y2_score`: `float32 [final_boxes_count, 7]` + + * NMS boxes + * final_boxes_count (N) ≠ batches + * 7 = [batch_no, classid, x1, y1, x2, y2, score] + +![image](https://github.com/PINTO0309/PINTO_model_zoo/assets/33194443/9d4fecdf-c90e-4e0a-99a5-9c3e61a4cf41) + +# How to generate post-processing ONNX +Simply change the following parameters and run all shells. + +https://github.com/PINTO0309/PINTO_model_zoo/blob/main/420_Gold-YOLO-Hand/post_process_gen_tools/convert_script.sh +```bash +OPSET=11 +BATCHES=1 +BOXES=5040 +CLASSES=1 +``` + +```bash +sudo chmod +x ./convert_script.sh +./convert_script.sh +``` + +# How to change NMS parameters +![image](https://user-images.githubusercontent.com/33194443/178084918-af33bfcc-425f-496e-87fb-1331ef7b2b6e.png) + +https://github.com/PINTO0309/simple-onnx-processing-tools + +Run the script below to directly rewrite the parameters of the ONNX file. +```bash +### Number of output boxes for Gold-YOLO +BOXES=5040 + +### max_output_boxes_per_class +sam4onnx \ +--op_name main01_nonmaxsuppression11 \ +--input_onnx_file_path 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--output_onnx_file_path 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--input_constants main01_max_output_boxes_per_class int64 [10] + +### iou_threshold +sam4onnx \ +--op_name main01_nonmaxsuppression11 \ +--input_onnx_file_path 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--output_onnx_file_path 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--input_constants main01_iou_threshold float32 [0.05] + +### score_threshold +sam4onnx \ +--op_name main01_nonmaxsuppression11 \ +--input_onnx_file_path 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--output_onnx_file_path 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--input_constants main01_score_threshold float32 [0.25] +``` + +# How to merge post-processing into a Gold-YOLO model +Simply execute the following command. + +https://github.com/PINTO0309/simple-onnx-processing-tools + +```bash +################################################### Gold-YOLO + Post-Process +MODEL=gold_yolo +BOXES=5040 +H=256 +W=320 + +snc4onnx \ +--input_onnx_file_paths ${MODEL}_${H}x${W}.onnx 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx + +################################################### 1 Batch + +MODEL=gold_yolo + +BOXES=5040 +H=256 +W=320 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_${H}x${W}.onnx 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx + +BOXES=7560 +H=256 +W=480 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_${H}x${W}.onnx 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx + +BOXES=10080 +H=256 +W=640 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_${H}x${W}.onnx 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx + +BOXES=15120 +H=384 +W=640 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_${H}x${W}.onnx 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx + +BOXES=18900 +H=480 +W=640 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_${H}x${W}.onnx 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx + +BOXES=57960 +H=736 +W=1280 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_${H}x${W}.onnx 30_nms_gold_yolo_m_hand_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx +onnxsim ${MODEL}_post_${H}x${W}.onnx ${MODEL}_post_${H}x${W}.onnx + +################################################### N Batch + +MODEL=gold_yolo + +BOXES=5040 +H=256 +W=320 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_Nx3x${H}x${W}.onnx 31_nms_gold_yolo_m_hand_N_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx + +BOXES=7560 +H=256 +W=480 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_Nx3x${H}x${W}.onnx 31_nms_gold_yolo_m_hand_N_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx + +BOXES=10080 +H=256 +W=640 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_Nx3x${H}x${W}.onnx 31_nms_gold_yolo_m_hand_N_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx + +BOXES=15120 +H=384 +W=640 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_Nx3x${H}x${W}.onnx 31_nms_gold_yolo_m_hand_N_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx + +BOXES=18900 +H=480 +W=640 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_Nx3x${H}x${W}.onnx 31_nms_gold_yolo_m_hand_N_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx + +BOXES=57960 +H=736 +W=1280 +snc4onnx \ +--input_onnx_file_paths ${MODEL}_Nx3x${H}x${W}.onnx 31_nms_gold_yolo_m_hand_N_${BOXES}.onnx \ +--srcop_destop output predictions \ +--output_onnx_file_path ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx +onnxsim ${MODEL}_post_Nx3x${H}x${W}.onnx ${MODEL}_post_Nx3x${H}x${W}.onnx +``` diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/boxes_listup.py b/421_Gold-YOLO-Head/post_process_gen_tools/boxes_listup.py new file mode 100644 index 0000000000..4d6be5b33f --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/boxes_listup.py @@ -0,0 +1,54 @@ +import onnxruntime as ort +from collections import OrderedDict + +MODELS = OrderedDict( + { + "01": "gold_yolo_n_hand_0423_0.2295_1x3x192x320.onnx", + "02": "gold_yolo_n_hand_0423_0.2295_1x3x192x416.onnx", + "03": "gold_yolo_n_hand_0423_0.2295_1x3x192x640.onnx", + "04": "gold_yolo_n_hand_0423_0.2295_1x3x192x800.onnx", + "05": "gold_yolo_n_hand_0423_0.2295_1x3x256x320.onnx", + "06": "gold_yolo_n_hand_0423_0.2295_1x3x256x416.onnx", + "07": "gold_yolo_n_hand_0423_0.2295_1x3x256x640.onnx", + "08": "gold_yolo_n_hand_0423_0.2295_1x3x256x800.onnx", + "09": "gold_yolo_n_hand_0423_0.2295_1x3x256x960.onnx", + "10": "gold_yolo_n_hand_0423_0.2295_1x3x288x1280.onnx", + "11": "gold_yolo_n_hand_0423_0.2295_1x3x288x480.onnx", + "12": "gold_yolo_n_hand_0423_0.2295_1x3x288x640.onnx", + "13": "gold_yolo_n_hand_0423_0.2295_1x3x288x800.onnx", + "14": "gold_yolo_n_hand_0423_0.2295_1x3x288x960.onnx", + "15": "gold_yolo_n_hand_0423_0.2295_1x3x320x320.onnx", + "16": "gold_yolo_n_hand_0423_0.2295_1x3x384x1280.onnx", + "17": "gold_yolo_n_hand_0423_0.2295_1x3x384x480.onnx", + "18": "gold_yolo_n_hand_0423_0.2295_1x3x384x640.onnx", + "19": "gold_yolo_n_hand_0423_0.2295_1x3x384x800.onnx", + "20": "gold_yolo_n_hand_0423_0.2295_1x3x384x960.onnx", + "21": "gold_yolo_n_hand_0423_0.2295_1x3x416x416.onnx", + "22": "gold_yolo_n_hand_0423_0.2295_1x3x480x1280.onnx", + "23": "gold_yolo_n_hand_0423_0.2295_1x3x480x640.onnx", + "24": "gold_yolo_n_hand_0423_0.2295_1x3x480x800.onnx", + "25": "gold_yolo_n_hand_0423_0.2295_1x3x480x960.onnx", + "26": "gold_yolo_n_hand_0423_0.2295_1x3x512x512.onnx", + "27": "gold_yolo_n_hand_0423_0.2295_1x3x512x640.onnx", + "28": "gold_yolo_n_hand_0423_0.2295_1x3x512x896.onnx", + "29": "gold_yolo_n_hand_0423_0.2295_1x3x544x1280.onnx", + "30": "gold_yolo_n_hand_0423_0.2295_1x3x544x800.onnx", + "31": "gold_yolo_n_hand_0423_0.2295_1x3x544x960.onnx", + "32": "gold_yolo_n_hand_0423_0.2295_1x3x640x640.onnx", + "33": "gold_yolo_n_hand_0423_0.2295_1x3x736x1280.onnx", + } +) + +box_sizes = [] +for k, v in MODELS.items(): + onnx_session = ort.InferenceSession( + path_or_bytes=v, + providers=['CPUExecutionProvider'], + ) + box_sizes.append([onnx_session.get_inputs()[0].shape[2], onnx_session.get_inputs()[0].shape[3], onnx_session.get_outputs()[0].shape[1]]) + +print(f'MODELS count: {len(MODELS)}') +print(f'BOX_SIZE count: {len(box_sizes)}') + +for box_size in box_sizes: + print(f'"{box_size[0]} {box_size[1]} {box_size[2]}"') \ No newline at end of file diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/convert_script.sh b/421_Gold-YOLO-Head/post_process_gen_tools/convert_script.sh new file mode 100755 index 0000000000..17d43dd17e --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/convert_script.sh @@ -0,0 +1,484 @@ +# pip install -U pip \ +# && pip install onnxsim==0.4.33 \ +# && pip install -U simple-onnx-processing-tools \ +# && pip install -U onnx \ +# && python3 -m pip install -U onnx_graphsurgeon --index-url https://pypi.ngc.nvidia.com + +MODEL_NAME=gold_yolo_n_hand +SUFFIX="0333_0.4040_1x3x" +OPSET=11 +BATCHES=1 +CLASSES=1 + +RESOLUTIONS=( + "192 320 1260" + "192 416 1638" + "192 640 2520" + "192 800 3150" + "256 320 1680" + "256 416 2184" + "256 640 3360" + "256 800 4200" + "256 960 5040" + "288 1280 7560" + "288 480 2835" + "288 640 3780" + "288 800 4725" + "288 960 5670" + "320 320 2100" + "384 1280 10080" + "384 480 3780" + "384 640 5040" + "384 800 6300" + "384 960 7560" + "416 416 3549" + "480 1280 12600" + "480 640 6300" + "480 800 7875" + "480 960 9450" + "512 512 5376" + "512 640 6720" + "512 896 9408" + "544 1280 14280" + "544 800 8925" + "544 960 10710" + "640 640 8400" + "736 1280 19320" +) + +for((i=0; i<${#RESOLUTIONS[@]}; i++)) +do + RESOLUTION=(`echo ${RESOLUTIONS[i]}`) + H=${RESOLUTION[0]} + W=${RESOLUTION[1]} + BOXES=${RESOLUTION[2]} + + ################################################### Boxes + Scores + python make_boxes_scores.py -o ${OPSET} -b ${BATCHES} -x ${BOXES} -c ${CLASSES} + python make_cxcywh_y1x1y2x2.py -o ${OPSET} -b ${BATCHES} -x ${BOXES} + + sor4onnx \ + --input_onnx_file_path 01_boxes_scores_${BOXES}.onnx \ + --old_new "/Constant" "boxes_scores_Constant" \ + --mode full \ + --search_mode prefix_match \ + --output_onnx_file_path 01_boxes_scores_${BOXES}.onnx + + sor4onnx \ + --input_onnx_file_path 02_cxcywh_y1x1y2x2_${BOXES}.onnx \ + --old_new "/Constant" "cxcywh_y1x1y2x2_Constant" \ + --mode full \ + --search_mode prefix_match \ + --output_onnx_file_path 02_cxcywh_y1x1y2x2_${BOXES}.onnx + + sor4onnx \ + --input_onnx_file_path 02_cxcywh_y1x1y2x2_${BOXES}.onnx \ + --old_new "/Slice" "cxcywh_y1x1y2x2_Slice" \ + --mode full \ + --search_mode prefix_match \ + --output_onnx_file_path 02_cxcywh_y1x1y2x2_${BOXES}.onnx + + snc4onnx \ + --input_onnx_file_paths 01_boxes_scores_${BOXES}.onnx 02_cxcywh_y1x1y2x2_${BOXES}.onnx \ + --srcop_destop boxes_cxcywh cxcywh \ + --output_onnx_file_path 03_boxes_y1x1y2x2_scores_${BOXES}.onnx + + + ################################################### NonMaxSuppression + sog4onnx \ + --op_type Constant \ + --opset ${OPSET} \ + --op_name max_output_boxes_per_class_const \ + --output_variables max_output_boxes_per_class int64 [1] \ + --attributes value int64 [20] \ + --output_onnx_file_path 04_Constant_max_output_boxes_per_class.onnx + + # N: iou_threshold_const=0.40, score_threshold_const=0.25 + # M: iou_threshold_const=0.40, score_threshold_const=0.25 + + sog4onnx \ + --op_type Constant \ + --opset ${OPSET} \ + --op_name iou_threshold_const \ + --output_variables iou_threshold float32 [1] \ + --attributes value float32 [0.40] \ + --output_onnx_file_path 05_Constant_iou_threshold.onnx + + sog4onnx \ + --op_type Constant \ + --opset ${OPSET} \ + --op_name score_threshold_const \ + --output_variables score_threshold float32 [1] \ + --attributes value float32 [0.25] \ + --output_onnx_file_path 06_Constant_score_threshold.onnx + + + OP=NonMaxSuppression + LOWEROP=${OP,,} + sog4onnx \ + --op_type ${OP} \ + --opset ${OPSET} \ + --op_name ${LOWEROP}${OPSET} \ + --input_variables boxes_var float32 [${BATCHES},${BOXES},4] \ + --input_variables scores_var float32 [${BATCHES},${CLASSES},${BOXES}] \ + --input_variables max_output_boxes_per_class_var int64 [1] \ + --input_variables iou_threshold_var float32 [1] \ + --input_variables score_threshold_var float32 [1] \ + --output_variables selected_indices int64 [\'N\',3] \ + --attributes center_point_box int64 0 \ + --output_onnx_file_path 07_${OP}${OPSET}.onnx + + + snc4onnx \ + --input_onnx_file_paths 04_Constant_max_output_boxes_per_class.onnx 07_${OP}${OPSET}.onnx \ + --srcop_destop max_output_boxes_per_class max_output_boxes_per_class_var \ + --output_onnx_file_path 07_${OP}${OPSET}.onnx + + snc4onnx \ + --input_onnx_file_paths 05_Constant_iou_threshold.onnx 07_${OP}${OPSET}.onnx \ + --srcop_destop iou_threshold iou_threshold_var \ + --output_onnx_file_path 07_${OP}${OPSET}.onnx + + snc4onnx \ + --input_onnx_file_paths 06_Constant_score_threshold.onnx 07_${OP}${OPSET}.onnx \ + --srcop_destop score_threshold score_threshold_var \ + --output_onnx_file_path 07_${OP}${OPSET}.onnx + + soc4onnx \ + --input_onnx_file_path 07_${OP}${OPSET}.onnx \ + --output_onnx_file_path 07_${OP}${OPSET}.onnx \ + --opset ${OPSET} + + + ################################################### Boxes + Scores + NonMaxSuppression + snc4onnx \ + --input_onnx_file_paths 03_boxes_y1x1y2x2_scores_${BOXES}.onnx 07_${OP}${OPSET}.onnx \ + --srcop_destop scores scores_var y1x1y2x2 boxes_var \ + --output_onnx_file_path 08_nms_${MODEL_NAME}_${BOXES}.onnx + + + ################################################### Myriad workaround Mul + OP=Mul + LOWEROP=${OP,,} + OPSET=${OPSET} + sog4onnx \ + --op_type ${OP} \ + --opset ${OPSET} \ + --op_name ${LOWEROP}${OPSET} \ + --input_variables workaround_mul_a int64 [\'N\',3] \ + --input_variables workaround_mul_b int64 [1] \ + --output_variables workaround_mul_out int64 [\'N\',3] \ + --output_onnx_file_path 09_${OP}${OPSET}_workaround.onnx + + ############ Myriad workaround Constant + sog4onnx \ + --op_type Constant \ + --opset ${OPSET} \ + --op_name workaround_mul_const_op \ + --output_variables workaround_mul_const int64 [1] \ + --attributes value int64 [1] \ + --output_onnx_file_path 10_Constant_workaround_mul.onnx + + ############ Myriad workaround Mul + Myriad workaround Constant + snc4onnx \ + --input_onnx_file_paths 10_Constant_workaround_mul.onnx 09_${OP}${OPSET}_workaround.onnx \ + --srcop_destop workaround_mul_const workaround_mul_b \ + --output_onnx_file_path 09_${OP}${OPSET}_workaround.onnx + + + + ################################################### NonMaxSuppression + Myriad workaround Mul + snc4onnx \ + --input_onnx_file_paths 08_nms_${MODEL_NAME}_${BOXES}.onnx 09_${OP}${OPSET}_workaround.onnx \ + --srcop_destop selected_indices workaround_mul_a \ + --output_onnx_file_path 11_nms_${MODEL_NAME}_${BOXES}.onnx \ + --disable_onnxsim + + ################################################### N batch NMS + sbi4onnx \ + --input_onnx_file_path 11_nms_${MODEL_NAME}_${BOXES}.onnx \ + --output_onnx_file_path 12_nms_${MODEL_NAME}_${BOXES}_batch.onnx \ + --initialization_character_string batch + + sio4onnx \ + --input_onnx_file_path 12_nms_${MODEL_NAME}_${BOXES}_batch.onnx \ + --output_onnx_file_path 12_nms_${MODEL_NAME}_${BOXES}_batch.onnx \ + --input_names "predictions" \ + --input_shapes "batch" ${BOXES} $((CLASSES+5)) \ + --output_names "x1y1x2y2" \ + --output_names "workaround_mul_out" \ + --output_shapes "batch" ${BOXES} 4 \ + --output_shapes "N" 3 + + + + + ################################################### Score GatherND + python make_score_gather_nd.py -b ${BATCHES} -x ${BOXES} -c ${CLASSES} + + python -m tf2onnx.convert \ + --opset ${OPSET} \ + --tflite saved_model_postprocess/nms_score_gather_nd.tflite \ + --output 13_nms_score_gather_nd.onnx + + sor4onnx \ + --input_onnx_file_path 13_nms_score_gather_nd.onnx \ + --old_new ":0" "" \ + --mode full \ + --search_mode partial_match \ + --output_onnx_file_path 13_nms_score_gather_nd.onnx + + sor4onnx \ + --input_onnx_file_path 13_nms_score_gather_nd.onnx \ + --old_new "serving_default_input_1" "gn_scores" \ + --output_onnx_file_path 13_nms_score_gather_nd.onnx \ + --mode inputs + + sor4onnx \ + --input_onnx_file_path 13_nms_score_gather_nd.onnx \ + --old_new "serving_default_input_2" "gn_selected_indices" \ + --output_onnx_file_path 13_nms_score_gather_nd.onnx \ + --mode inputs + + sor4onnx \ + --input_onnx_file_path 13_nms_score_gather_nd.onnx \ + --old_new "PartitionedCall" "final_scores" \ + --output_onnx_file_path 13_nms_score_gather_nd.onnx \ + --mode outputs + + python make_input_output_shape_update.py \ + --input_onnx_file_path 13_nms_score_gather_nd.onnx \ + --output_onnx_file_path 13_nms_score_gather_nd.onnx \ + --input_names gn_scores \ + --input_names gn_selected_indices \ + --input_shapes ${BATCHES} ${CLASSES} ${BOXES} \ + --input_shapes N 3 \ + --output_names final_scores \ + --output_shapes N 1 + + onnxsim 13_nms_score_gather_nd.onnx 13_nms_score_gather_nd.onnx + onnxsim 13_nms_score_gather_nd.onnx 13_nms_score_gather_nd.onnx + + sio4onnx \ + --input_onnx_file_path 13_nms_score_gather_nd.onnx \ + --output_onnx_file_path 14_nms_score_gather_nd_batch.onnx \ + --input_names "gn_scores" \ + --input_names "gn_selected_indices" \ + --input_shapes "batch" ${CLASSES} ${BOXES} \ + --input_shapes "N" 3 \ + --output_names "final_scores" \ + --output_shapes "N" 1 + + + ################################################### NonMaxSuppression + Score GatherND + snc4onnx \ + --input_onnx_file_paths 11_nms_${MODEL_NAME}_${BOXES}.onnx 13_nms_score_gather_nd.onnx \ + --srcop_destop scores gn_scores workaround_mul_out gn_selected_indices \ + --output_onnx_file_path 15_nms_${MODEL_NAME}_${BOXES}_nd.onnx + + onnxsim 15_nms_${MODEL_NAME}_${BOXES}_nd.onnx 15_nms_${MODEL_NAME}_${BOXES}_nd.onnx + onnxsim 15_nms_${MODEL_NAME}_${BOXES}_nd.onnx 15_nms_${MODEL_NAME}_${BOXES}_nd.onnx + + + snc4onnx \ + --input_onnx_file_paths 12_nms_${MODEL_NAME}_${BOXES}_batch.onnx 14_nms_score_gather_nd_batch.onnx \ + --srcop_destop scores gn_scores workaround_mul_out gn_selected_indices \ + --output_onnx_file_path 16_nms_${MODEL_NAME}_${BOXES}_nd_batch.onnx + + onnxsim 16_nms_${MODEL_NAME}_${BOXES}_nd_batch.onnx 16_nms_${MODEL_NAME}_${BOXES}_nd_batch.onnx + onnxsim 16_nms_${MODEL_NAME}_${BOXES}_nd_batch.onnx 16_nms_${MODEL_NAME}_${BOXES}_nd_batch.onnx + + + + + + + + ################################################### Final Batch Nums + python make_final_batch_nums_final_class_nums_final_box_nums.py + + + ################################################### Boxes GatherND + python make_box_gather_nd.py + + python -m tf2onnx.convert \ + --opset ${OPSET} \ + --tflite saved_model_postprocess/nms_box_gather_nd.tflite \ + --output 18_nms_box_gather_nd.onnx + + sor4onnx \ + --input_onnx_file_path 18_nms_box_gather_nd.onnx \ + --old_new ":0" "" \ + --mode full \ + --search_mode partial_match \ + --output_onnx_file_path 18_nms_box_gather_nd.onnx + + sor4onnx \ + --input_onnx_file_path 18_nms_box_gather_nd.onnx \ + --old_new "serving_default_input_1" "gn_boxes" \ + --output_onnx_file_path 18_nms_box_gather_nd.onnx \ + --mode inputs + + sor4onnx \ + --input_onnx_file_path 18_nms_box_gather_nd.onnx \ + --old_new "serving_default_input_2" "gn_box_selected_indices" \ + --output_onnx_file_path 18_nms_box_gather_nd.onnx \ + --mode inputs + + sor4onnx \ + --input_onnx_file_path 18_nms_box_gather_nd.onnx \ + --old_new "PartitionedCall" "final_boxes" \ + --output_onnx_file_path 18_nms_box_gather_nd.onnx \ + --mode outputs + + python make_input_output_shape_update.py \ + --input_onnx_file_path 18_nms_box_gather_nd.onnx \ + --output_onnx_file_path 18_nms_box_gather_nd.onnx \ + --input_names gn_boxes \ + --input_names gn_box_selected_indices \ + --input_shapes ${BATCHES} ${BOXES} 4 \ + --input_shapes N 2 \ + --output_names final_boxes \ + --output_shapes N 4 + + onnxsim 18_nms_box_gather_nd.onnx 18_nms_box_gather_nd.onnx + onnxsim 18_nms_box_gather_nd.onnx 18_nms_box_gather_nd.onnx + + sio4onnx \ + --input_onnx_file_path 18_nms_box_gather_nd.onnx \ + --output_onnx_file_path 19_nms_box_gather_nd_batch.onnx \ + --input_names "gn_boxes" \ + --input_names "gn_box_selected_indices" \ + --input_shapes "batch" ${BOXES} 4 \ + --input_shapes "N" 2 \ + --output_names "final_boxes" \ + --output_shapes "N" 4 + + + ################################################### nms_${MODEL_NAME}_xxx_nd + nms_final_batch_nums_final_class_nums_final_box_nums + snc4onnx \ + --input_onnx_file_paths 15_nms_${MODEL_NAME}_${BOXES}_nd.onnx 17_nms_final_batch_nums_final_class_nums_final_box_nums.onnx \ + --srcop_destop selected_indices bc_input \ + --op_prefixes_after_merging main01 sub01 \ + --output_onnx_file_path 20_nms_${MODEL_NAME}_${BOXES}_split.onnx + + snc4onnx \ + --input_onnx_file_paths 16_nms_${MODEL_NAME}_${BOXES}_nd_batch.onnx 17_nms_final_batch_nums_final_class_nums_final_box_nums.onnx \ + --srcop_destop workaround_mul_out bc_input \ + --op_prefixes_after_merging main01 sub01 \ + --output_onnx_file_path 21_nms_${MODEL_NAME}_${BOXES}_split_batch.onnx + + + + ################################################### nms_${MODEL_NAME}_${BOXES}_split + nms_box_gather_nd + snc4onnx \ + --input_onnx_file_paths 20_nms_${MODEL_NAME}_${BOXES}_split.onnx 18_nms_box_gather_nd.onnx \ + --srcop_destop x1y1x2y2 gn_boxes final_box_nums gn_box_selected_indices \ + --output_onnx_file_path 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx + + onnxsim 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx + onnxsim 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx + + + snc4onnx \ + --input_onnx_file_paths 21_nms_${MODEL_NAME}_${BOXES}_split_batch.onnx 19_nms_box_gather_nd_batch.onnx \ + --srcop_destop x1y1x2y2 gn_boxes final_box_nums gn_box_selected_indices \ + --output_onnx_file_path 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx + + onnxsim 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx + onnxsim 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx + + + + + + ################################################### nms output op name Cleaning + sor4onnx \ + --input_onnx_file_path 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx \ + --old_new "main01_final_scores" "final_scores" \ + --output_onnx_file_path 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx \ + --mode outputs + + sor4onnx \ + --input_onnx_file_path 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx \ + --old_new "sub01_final_batch_nums" "final_batch_nums" \ + --output_onnx_file_path 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx \ + --mode outputs + + sor4onnx \ + --input_onnx_file_path 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx \ + --old_new "sub01_final_class_nums" "final_class_nums" \ + --output_onnx_file_path 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx \ + --mode outputs + + + sor4onnx \ + --input_onnx_file_path 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx \ + --old_new "main01_final_scores" "final_scores" \ + --output_onnx_file_path 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx \ + --mode outputs + + sor4onnx \ + --input_onnx_file_path 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx \ + --old_new "sub01_final_batch_nums" "final_batch_nums" \ + --output_onnx_file_path 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx \ + --mode outputs + + sor4onnx \ + --input_onnx_file_path 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx \ + --old_new "sub01_final_class_nums" "final_class_nums" \ + --output_onnx_file_path 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx \ + --mode outputs + + + + + + + + ################################################### nms output merge + python make_nms_outputs_merge.py + + onnxsim 24_nms_batchno_classid_x1y1x2y2_score_cat.onnx 24_nms_batchno_classid_x1y1x2y2_score_cat.onnx + + + ################################################### merge + snc4onnx \ + --input_onnx_file_paths 22_nms_${MODEL_NAME}_${BOXES}_merged.onnx 24_nms_batchno_classid_x1y1x2y2_score_cat.onnx \ + --srcop_destop final_batch_nums cat_batch final_class_nums cat_classid final_boxes cat_x1y1x2y2 final_scores cat_score \ + --output_onnx_file_path 30_nms_${MODEL_NAME}_${BOXES}.onnx + + sor4onnx \ + --input_onnx_file_path 30_nms_${MODEL_NAME}_${BOXES}.onnx \ + --old_new "final_scores" "score" \ + --output_onnx_file_path 30_nms_${MODEL_NAME}_${BOXES}.onnx \ + --mode outputs + + + + ################################################### merge + snc4onnx \ + --input_onnx_file_paths 23_nms_${MODEL_NAME}_${BOXES}_merged_batch.onnx 24_nms_batchno_classid_x1y1x2y2_score_cat.onnx \ + --srcop_destop final_batch_nums cat_batch final_class_nums cat_classid final_boxes cat_x1y1x2y2 final_scores cat_score \ + --output_onnx_file_path 31_nms_${MODEL_NAME}_N_${BOXES}.onnx + + sor4onnx \ + --input_onnx_file_path 31_nms_${MODEL_NAME}_N_${BOXES}.onnx \ + --old_new "final_scores" "score" \ + --output_onnx_file_path 31_nms_${MODEL_NAME}_N_${BOXES}.onnx \ + --mode outputs + + # ################################################### Cleaning + rm 0*.onnx + rm 1*.onnx + rm 2*.onnx + + + ################################################### ${MODEL_NAME} + Post-Process + snc4onnx \ + --input_onnx_file_paths ${MODEL_NAME}_${SUFFIX}${H}x${W}.onnx 30_nms_${MODEL_NAME}_${BOXES}.onnx \ + --srcop_destop output predictions \ + --output_onnx_file_path ${MODEL_NAME}_post_${SUFFIX}${H}x${W}.onnx + onnxsim ${MODEL_NAME}_post_${SUFFIX}${H}x${W}.onnx ${MODEL_NAME}_post_${SUFFIX}${H}x${W}.onnx + onnxsim ${MODEL_NAME}_post_${SUFFIX}${H}x${W}.onnx ${MODEL_NAME}_post_${SUFFIX}${H}x${W}.onnx +done diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/demo_goldyolo_onnx.py b/421_Gold-YOLO-Head/post_process_gen_tools/demo_goldyolo_onnx.py new file mode 100644 index 0000000000..901db733c2 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/demo_goldyolo_onnx.py @@ -0,0 +1,340 @@ +#!/usr/bin/env python + +import copy +import cv2 +import time +import numpy as np +import onnxruntime +from argparse import ArgumentParser +from typing import Tuple, Optional, List + + +class GoldYOLOONNX(object): + def __init__( + self, + model_path: Optional[str] = 'gold_yolo_m_hand_post_0465_0.2501_1x3x480x640.onnx', + class_score_th: Optional[float] = 0.40, + providers: Optional[List] = [ + ( + 'TensorrtExecutionProvider', { + 'trt_engine_cache_enable': True, + 'trt_engine_cache_path': '.', + 'trt_fp16_enable': True, + } + ), + 'CUDAExecutionProvider', + 'CPUExecutionProvider', + ], + ): + """YOLOv7ONNX + + Parameters + ---------- + model_path: Optional[str] + ONNX file path for YOLOv7 + + class_score_th: Optional[float] + Score threshold. Default: 0.25 + + providers: Optional[List] + Name of onnx execution providers + Default: + [ + ( + 'TensorrtExecutionProvider', { + 'trt_engine_cache_enable': True, + 'trt_engine_cache_path': '.', + 'trt_fp16_enable': True, + } + ), + 'CUDAExecutionProvider', + 'CPUExecutionProvider', + ] + """ + # Threshold + self.class_score_th = class_score_th + + # Model loading + session_option = onnxruntime.SessionOptions() + session_option.log_severity_level = 3 + self.onnx_session = onnxruntime.InferenceSession( + model_path, + sess_options=session_option, + providers=providers, + ) + self.providers = self.onnx_session.get_providers() + + self.input_shapes = [ + input.shape for input in self.onnx_session.get_inputs() + ] + self.input_names = [ + input.name for input in self.onnx_session.get_inputs() + ] + self.output_names = [ + output.name for output in self.onnx_session.get_outputs() + ] + + + def __call__( + self, + image: np.ndarray, + ) -> Tuple[np.ndarray, np.ndarray]: + """YOLOv7ONNX + + Parameters + ---------- + image: np.ndarray + Entire image + + Returns + ------- + boxes: np.ndarray + Predicted boxes: [N, y1, x1, y2, x2] + + scores: np.ndarray + Predicted box scores: [N, score] + """ + temp_image = copy.deepcopy(image) + + # PreProcess + resized_image = self.__preprocess( + temp_image, + ) + + # Inference + inferece_image = np.asarray([resized_image], dtype=np.float32) + boxes = self.onnx_session.run( + self.output_names, + {input_name: inferece_image for input_name in self.input_names}, + )[0] + + # PostProcess + result_boxes, result_scores = \ + self.__postprocess( + image=temp_image, + boxes=boxes, + ) + + return result_boxes, result_scores + + + def __preprocess( + self, + image: np.ndarray, + swap: Optional[Tuple[int,int,int]] = (2, 0, 1), + ) -> np.ndarray: + """__preprocess + + Parameters + ---------- + image: np.ndarray + Entire image + + swap: tuple + HWC to CHW: (2,0,1) + CHW to HWC: (1,2,0) + HWC to HWC: (0,1,2) + CHW to CHW: (0,1,2) + + Returns + ------- + resized_image: np.ndarray + Resized and normalized image. + """ + # Normalization + BGR->RGB + resized_image = cv2.resize( + image, + ( + int(self.input_shapes[0][3]), + int(self.input_shapes[0][2]), + ) + ) + resized_image = np.divide(resized_image, 255.0) + resized_image = resized_image[..., ::-1] + resized_image = resized_image.transpose(swap) + resized_image = np.ascontiguousarray( + resized_image, + dtype=np.float32, + ) + return resized_image + + + def __postprocess( + self, + image: np.ndarray, + boxes: np.ndarray, + ) -> Tuple[np.ndarray, np.ndarray]: + """__postprocess + + Parameters + ---------- + image: np.ndarray + Entire image. + + boxes: np.ndarray + float32[N, 7] + + Returns + ------- + result_boxes: np.ndarray + Predicted boxes: [N, y1, x1, y2, x2] + + result_scores: np.ndarray + Predicted box confs: [N, score] + """ + image_height = image.shape[0] + image_width = image.shape[1] + + """ + Detector is + N -> Number of boxes detected + batchno -> always 0: BatchNo.0 + + batchno_classid_y1x1y2x2_score: float32[N,7] + """ + result_boxes = [] + result_scores = [] + if len(boxes) > 0: + scores = boxes[:, 6:7] + keep_idxs = scores[:, 0] > self.class_score_th + scores_keep = scores[keep_idxs, :] + boxes_keep = boxes[keep_idxs, :] + + if len(boxes_keep) > 0: + for box, score in zip(boxes_keep, scores_keep): + x_min = max(int(box[2]), 0) + y_min = max(int(box[3]), 0) + x_max = min(int(box[4]), image_width) + y_max = min(int(box[5]), image_height) + + result_boxes.append( + [x_min, y_min, x_max, y_max] + ) + result_scores.append( + score + ) + + return np.asarray(result_boxes), np.asarray(result_scores) + + +def main(): + parser = ArgumentParser() + parser.add_argument( + '-m', + '--model', + type=str, + default='gold_yolo_m_hand_post_0465_0.2501_1x3x480x640.onnx', + ) + parser.add_argument( + '-v', + '--video', + type=int, + default=0, + ) + args = parser.parse_args() + + model = GoldYOLOONNX( + model_path=args.model, + ) + + cap = cv2.VideoCapture(args.video) + cap_fps = cap.get(cv2.CAP_PROP_FPS) + w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) + h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) + fourcc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v') + video_writer = cv2.VideoWriter( + filename='output.mp4', + fourcc=fourcc, + fps=cap_fps, + frameSize=(w, h), + ) + + while cap.isOpened(): + res, image = cap.read() + if not res: + break + + debug_image = copy.deepcopy(image) + + start_time = time.time() + boxes, scores = model(debug_image) + elapsed_time = time.time() - start_time + fps = 1 / elapsed_time + cv2.putText( + debug_image, + f'{fps:.1f} FPS (inferece + post-process)', + (10, 30), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (255, 255, 255), + 2, + cv2.LINE_AA, + ) + cv2.putText( + debug_image, + f'{fps:.1f} FPS (inferece + post-process)', + (10, 30), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (0, 0, 255), + 1, + cv2.LINE_AA, + ) + + for box, score in zip(boxes, scores): + cv2.rectangle( + debug_image, + (box[0], box[1]), + (box[2], box[3]), + (255,255,255), + 2, + ) + cv2.rectangle( + debug_image, + (box[0], box[1]), + (box[2], box[3]), + (0,0,255), + 1, + ) + cv2.putText( + debug_image, + f'{score[0]:.2f}', + ( + box[0], + box[1]-10 if box[1]-10 > 0 else 10 + ), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (255, 255, 255), + 2, + cv2.LINE_AA, + ) + cv2.putText( + debug_image, + f'{score[0]:.2f}', + ( + box[0], + box[1]-10 if box[1]-10 > 0 else 10 + ), + cv2.FONT_HERSHEY_SIMPLEX, + 0.7, + (0, 0, 255), + 1, + cv2.LINE_AA, + ) + + key = cv2.waitKey(1) + if key == 27: # ESC + break + + cv2.imshow("test", debug_image) + video_writer.write(debug_image) + + if video_writer: + video_writer.release() + + if cap: + cap.release() + +if __name__ == "__main__": + main() diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_batch_initialize.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_batch_initialize.py new file mode 100644 index 0000000000..50c6ddc6d8 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_batch_initialize.py @@ -0,0 +1,209 @@ +#! /usr/bin/env python + +import sys +import onnx +import onnx_graphsurgeon as gs +from typing import Optional +import struct +from argparse import ArgumentParser +from onnxsim import simplify + +class Color: + BLACK = '\033[30m' + RED = '\033[31m' + GREEN = '\033[32m' + YELLOW = '\033[33m' + BLUE = '\033[34m' + MAGENTA = '\033[35m' + CYAN = '\033[36m' + WHITE = '\033[37m' + COLOR_DEFAULT = '\033[39m' + BOLD = '\033[1m' + UNDERLINE = '\033[4m' + INVISIBLE = '\033[08m' + REVERCE = '\033[07m' + BG_BLACK = '\033[40m' + BG_RED = '\033[41m' + BG_GREEN = '\033[42m' + BG_YELLOW = '\033[43m' + BG_BLUE = '\033[44m' + BG_MAGENTA = '\033[45m' + BG_CYAN = '\033[46m' + BG_WHITE = '\033[47m' + BG_DEFAULT = '\033[49m' + RESET = '\033[0m' + + +TARGET_INPUTS = [ + 'predictions', +] + +TARGET_VALUE_INFO = [ + 'main01_boxes_cxcywh', + 'main01_onnx::Mul_15', + 'main01_onnx::Mul_10', + + 'main01_onnx::Div_28', + 'main01_onnx::Add_30', + 'main01_onnx::Add_26', + 'main01_onnx::Unsqueeze_31', + 'main01_onnx::Concat_32', + + 'main01_onnx::Div_20', + 'main01_onnx::Add_22', + 'main01_onnx::Add_18', + 'main01_onnx::Unsqueeze_23', + 'main01_onnx::Concat_24', + + 'main01_onnx::Div_12', + 'main01_onnx::Sub_14', + 'main01_onnx::Sub_10', + 'main01_onnx::Unsqueeze_15', + 'main01_onnx::Concat_16', + + 'main01_onnx::Div_4', + 'main01_onnx::Sub_6', + 'main01_onnx::Sub_2', + 'main01_onnx::Unsqueeze_7', + 'main01_onnx::Concat_8', + + 'main01_y1x1y2x2', + 'main01_onnx::Transpose_16', + 'main01_scores', +] + +def initialize( + input_onnx_file_path: Optional[str] = '', + onnx_graph: Optional[onnx.ModelProto] = None, + output_onnx_file_path: Optional[str] = '', + initialization_character_string: Optional[str] = 'batch', + non_verbose: Optional[bool] = False, +) -> onnx.ModelProto: + """ + Parameters + ---------- + input_onnx_file_path: Optional[str] + Input onnx file path.\n\ + Either input_onnx_file_path or onnx_graph must be specified.\n\ + Default: '' + onnx_graph: Optional[onnx.ModelProto] + onnx.ModelProto.\n\ + Either input_onnx_file_path or onnx_graph must be specified.\n\ + onnx_graph If specified, ignore input_onnx_file_path and process onnx_graph. + output_onnx_file_path: Optional[str] + Output onnx file path. If not specified, no ONNX file is output.\n\ + Default: '' + initialization_character_string: Optional[str] + String to initialize batch size. "-1" or "N" or "xxx", etc...\n + Default: 'batch' + non_verbose: Optional[bool] + Do not show all information logs. Only error logs are displayed.\n\ + Default: False + Returns + ------- + changed_graph: onnx.ModelProto + Changed onnx ModelProto. + """ + + # Unspecified check for input_onnx_file_path and onnx_graph + if not input_onnx_file_path and not onnx_graph: + print( + f'{Color.RED}ERROR:{Color.RESET} '+ + f'One of input_onnx_file_path or onnx_graph must be specified.' + ) + sys.exit(1) + + if not initialization_character_string: + print( + f'{Color.RED}ERROR:{Color.RESET} '+ + f'The initialization_character_string cannot be empty.' + ) + sys.exit(1) + + # Loading Graphs + # onnx_graph If specified, onnx_graph is processed first + if not onnx_graph: + onnx_graph = onnx.load(input_onnx_file_path) + try: + onnx_graph, _ = simplify(onnx_graph) + except: + pass + graph = gs.import_onnx(onnx_graph) + graph.cleanup().toposort() + target_model = gs.export_onnx(graph) + target_graph = target_model.graph + + for node in target_graph.input: + if node.name in TARGET_INPUTS: + if len(node.type.tensor_type.shape.dim)>0: + node.type.tensor_type.shape.dim[0].dim_param = initialization_character_string + + target_value_info = [value_info for value_info in target_graph.value_info if value_info.name in TARGET_VALUE_INFO] + for tensor in target_value_info: + if len(tensor.type.tensor_type.shape.dim)>0: + tensor.type.tensor_type.shape.dim[0].dim_param = initialization_character_string + + + # infer_shapes + target_model = onnx.shape_inference.infer_shapes(target_model) + + # Save + if output_onnx_file_path: + onnx.save(target_model, output_onnx_file_path) + + if not non_verbose: + print(f'{Color.GREEN}INFO:{Color.RESET} Finish!') + + # Return + return target_model + + +def main(): + parser = ArgumentParser() + parser.add_argument( + '--input_onnx_file_path', + type=str, + required=True, + help='Input onnx file path.' + ) + parser.add_argument( + '--output_onnx_file_path', + type=str, + required=True, + help='Output onnx file path.' + ) + parser.add_argument( + '--initialization_character_string', + type=str, + default='batch', + help=\ + 'String to initialize batch size. "-1" or "N" or "xxx", etc... \n'+ + 'Default: \'batch\'' + ) + parser.add_argument( + '--non_verbose', + action='store_true', + help='Do not show all information logs. Only error logs are displayed.' + ) + args = parser.parse_args() + + input_onnx_file_path = args.input_onnx_file_path + output_onnx_file_path = args.output_onnx_file_path + initialization_character_string = args.initialization_character_string + non_verbose = args.non_verbose + + # Load + onnx_graph = onnx.load(input_onnx_file_path) + + # Batchsize change + changed_graph = initialize( + input_onnx_file_path=None, + onnx_graph=onnx_graph, + output_onnx_file_path=output_onnx_file_path, + initialization_character_string=initialization_character_string, + non_verbose=non_verbose, + ) + + +if __name__ == '__main__': + main() \ No newline at end of file diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_box_gather_nd.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_box_gather_nd.py new file mode 100644 index 0000000000..7fd64f78c0 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_box_gather_nd.py @@ -0,0 +1,62 @@ +import os +os.environ['CUDA_VISIBLE_DEVICES'] = '-1' +import tensorflow as tf +import numpy as np +np.random.seed(0) +from argparse import ArgumentParser + + +if __name__ == "__main__": + parser = ArgumentParser() + parser.add_argument( + '-b', + '--batches', + type=int, + default=1, + help='batch size' + ) + parser.add_argument( + '-x', + '--boxes', + type=int, + default=5040, + help='boxes' + ) + args = parser.parse_args() + BATCHES = args.batches + BOXES = args.boxes + + # Create a model + boxes = tf.keras.layers.Input( + shape=[ + BOXES, + 4, + ], + batch_size=BATCHES, + dtype=tf.float32, + ) + + selected_indices = tf.keras.layers.Input( + shape=[ + 2, + ], + dtype=tf.int64, + ) + + gathered_boxes = tf.gather_nd( + boxes, + selected_indices, + batch_dims=0, + ) + + model = tf.keras.models.Model(inputs=[boxes, selected_indices], outputs=[gathered_boxes]) + model.summary() + output_path = 'saved_model_postprocess' + tf.saved_model.save(model, output_path) + converter = tf.lite.TFLiteConverter.from_keras_model(model) + converter.target_spec.supported_ops = [ + tf.lite.OpsSet.TFLITE_BUILTINS, + tf.lite.OpsSet.SELECT_TF_OPS + ] + tflite_model = converter.convert() + open(f"{output_path}/nms_box_gather_nd.tflite", "wb").write(tflite_model) \ No newline at end of file diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_boxes_scores.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_boxes_scores.py new file mode 100644 index 0000000000..054cfa5ab2 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_boxes_scores.py @@ -0,0 +1,97 @@ +#! /usr/bin/env python + +import torch +import torch.nn as nn +import onnx +from onnxsim import simplify +from argparse import ArgumentParser + +""" +prediction [1, 5040, 85] + +80 classes + +85 + +[0] -> center_x +[1] -> center_y +[2] -> width +[3] -> height +[4] -> box_score +[5]-[84] -> class_score +""" + + +class Model(nn.Module): + def __init__(self): + super(Model, self).__init__() + + def forward(self, x): + boxes = x[..., :4] # xywh [n, boxes, 4] + box_scores = x[..., 4:5] # [n, boxes, 1] + class_scores = x[..., 5:] # [n, boxes, 80] + scores = box_scores * class_scores + # scores = torch.sqrt(scores) + scores = scores.permute(0,2,1) + return boxes, scores + + +if __name__ == "__main__": + parser = ArgumentParser() + parser.add_argument( + '-o', + '--opset', + type=int, + default=11, + help='onnx opset' + ) + parser.add_argument( + '-b', + '--batches', + type=int, + default=1, + help='batch size' + ) + parser.add_argument( + '-x', + '--boxes', + type=int, + default=5040, + help='boxes' + ) + parser.add_argument( + '-c', + '--classes', + type=int, + default=80, + help='classes' + ) + args = parser.parse_args() + + model = Model() + + MODEL = f'01_boxes_scores' + OPSET=args.opset + BATCHES = args.batches + BOXES = args.boxes + CLASSES = args.classes + + onnx_file = f"{MODEL}_{BOXES}.onnx" + x = torch.randn(BATCHES, BOXES, CLASSES+5) + + torch.onnx.export( + model, + args=(x), + f=onnx_file, + opset_version=OPSET, + input_names = ['predictions'], + output_names=['boxes_cxcywh','scores'], + ) + model_onnx1 = onnx.load(onnx_file) + model_onnx1 = onnx.shape_inference.infer_shapes(model_onnx1) + onnx.save(model_onnx1, onnx_file) + + model_onnx2 = onnx.load(onnx_file) + model_simp, check = simplify(model_onnx2) + onnx.save(model_simp, onnx_file) + diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_cxcywh_x1y1x2y2.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_cxcywh_x1y1x2y2.py new file mode 100644 index 0000000000..3c564b2466 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_cxcywh_x1y1x2y2.py @@ -0,0 +1,72 @@ +#! /usr/bin/env python + +import torch +import torch.nn as nn +import numpy as np +import onnx +from onnxsim import simplify +from argparse import ArgumentParser + +class Model(nn.Module): + def __init__(self): + super(Model, self).__init__() + + def forward(self, cxcywh): + x1 = (cxcywh[..., 0:1] - cxcywh[..., 2:3] / 2) # top left x + y1 = (cxcywh[..., 1:2] - cxcywh[..., 3:4] / 2) # top left y + x2 = (cxcywh[..., 0:1] + cxcywh[..., 2:3] / 2) # bottom right x + y2 = (cxcywh[..., 1:2] + cxcywh[..., 3:4] / 2) # bottom right y + x1y1x2y2 = torch.cat([x1,y1,x2,y2], dim=2) + return x1y1x2y2 + + +if __name__ == "__main__": + parser = ArgumentParser() + parser.add_argument( + '-o', + '--opset', + type=int, + default=11, + help='onnx opset' + ) + parser.add_argument( + '-b', + '--batches', + type=int, + default=1, + help='batch size' + ) + parser.add_argument( + '-x', + '--boxes', + type=int, + default=5040, + help='boxes' + ) + args = parser.parse_args() + + model = Model() + + MODEL = f'cxcywh_x1y1x2y2' + OPSET=args.opset + BATCHES = args.batches + BOXES = args.boxes + + onnx_file = f"{MODEL}_{BOXES}.onnx" + cxcywh = torch.randn(BATCHES, BOXES, 4) + + torch.onnx.export( + model, + args=(cxcywh), + f=onnx_file, + opset_version=OPSET, + input_names = ['cxcywh'], + output_names=['x1y1x2y2'], + ) + model_onnx1 = onnx.load(onnx_file) + model_onnx1 = onnx.shape_inference.infer_shapes(model_onnx1) + onnx.save(model_onnx1, onnx_file) + + model_onnx2 = onnx.load(onnx_file) + model_simp, check = simplify(model_onnx2) + onnx.save(model_simp, onnx_file) diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_cxcywh_y1x1y2x2.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_cxcywh_y1x1y2x2.py new file mode 100644 index 0000000000..705ac5b0e0 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_cxcywh_y1x1y2x2.py @@ -0,0 +1,78 @@ +#! /usr/bin/env python + +import torch +import torch.nn as nn +import numpy as np +import onnx +from onnxsim import simplify +from argparse import ArgumentParser + +class Model(nn.Module): + def __init__(self): + super(Model, self).__init__() + + def forward(self, cxcywh): + x1 = (cxcywh[..., 0:1] - cxcywh[..., 2:3] / 2) # top left x + y1 = (cxcywh[..., 1:2] - cxcywh[..., 3:4] / 2) # top left y + x2 = (cxcywh[..., 0:1] + cxcywh[..., 2:3] / 2) # bottom right x + y2 = (cxcywh[..., 1:2] + cxcywh[..., 3:4] / 2) # bottom right y + y1x1y2x2 = torch.cat([y1,x1,y2,x2], dim=2) + x1y1x2y2 = torch.cat([x1,y1,x2,y2], dim=2) + return y1x1y2x2, x1y1x2y2 + + +if __name__ == "__main__": + parser = ArgumentParser() + parser.add_argument( + '-o', + '--opset', + type=int, + default=11, + help='onnx opset' + ) + parser.add_argument( + '-b', + '--batches', + type=int, + default=1, + help='batch size' + ) + parser.add_argument( + '-x', + '--boxes', + type=int, + default=5040, + help='boxes' + ) + args = parser.parse_args() + + model = Model() + + MODEL = f'02_cxcywh_y1x1y2x2' + OPSET=args.opset + BATCHES = args.batches + BOXES = args.boxes + + onnx_file = f"{MODEL}_{BOXES}.onnx" + cxcywh = torch.randn(BATCHES, BOXES, 4) + + torch.onnx.export( + model, + args=(cxcywh), + f=onnx_file, + opset_version=OPSET, + input_names = [ + 'cxcywh', + ], + output_names=[ + 'y1x1y2x2', + 'x1y1x2y2', + ], + ) + model_onnx1 = onnx.load(onnx_file) + model_onnx1 = onnx.shape_inference.infer_shapes(model_onnx1) + onnx.save(model_onnx1, onnx_file) + + model_onnx2 = onnx.load(onnx_file) + model_simp, check = simplify(model_onnx2) + onnx.save(model_simp, onnx_file) diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_final_batch_nums_final_class_nums_final_box_nums.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_final_batch_nums_final_class_nums_final_box_nums.py new file mode 100644 index 0000000000..dace921053 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_final_batch_nums_final_class_nums_final_box_nums.py @@ -0,0 +1,61 @@ +#! /usr/bin/env python + +import torch +import torch.nn as nn +import onnx +import numpy as np +from onnxsim import simplify +from argparse import ArgumentParser + +class Model(nn.Module): + def __init__(self): + super(Model, self).__init__() + + def forward(self, x): + batch_nums = x[:, 0:1].to(torch.float32) # batch number + class_nums = x[:, 1:2].to(torch.float32) # class ids + box_nums = x[:, [0,2]] # batch number + box number + return batch_nums, class_nums, box_nums + + +if __name__ == "__main__": + parser = ArgumentParser() + parser.add_argument( + '-o', + '--opset', + type=int, + default=11, + help='onnx opset' + ) + args = parser.parse_args() + + model = Model() + + MODEL = f'17_nms_final_batch_nums_final_class_nums_final_box_nums' + OPSET=args.opset + + onnx_file = f"{MODEL}.onnx" + x = torch.ones([1, 3], dtype=torch.int64) + + torch.onnx.export( + model, + args=(x), + f=onnx_file, + opset_version=OPSET, + input_names=['bc_input'], + output_names=['final_batch_nums','final_class_nums','final_box_nums'], + dynamic_axes={ + 'bc_input': {0: 'N'}, + 'final_batch_nums': {0: 'N'}, + 'final_class_nums': {0: 'N'}, + 'final_box_nums': {0: 'N'}, + } + ) + model_onnx1 = onnx.load(onnx_file) + model_onnx1 = onnx.shape_inference.infer_shapes(model_onnx1) + onnx.save(model_onnx1, onnx_file) + + model_onnx2 = onnx.load(onnx_file) + model_simp, check = simplify(model_onnx2) + onnx.save(model_simp, onnx_file) + diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_input_output_shape_update.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_input_output_shape_update.py new file mode 100644 index 0000000000..f00d8fc5dc --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_input_output_shape_update.py @@ -0,0 +1,76 @@ +import onnx +from onnx.tools import update_model_dims +from argparse import ArgumentParser + + +if __name__ == "__main__": + parser = ArgumentParser() + parser.add_argument( + '-if', + '--input_onnx_file_path', + type=str, + required=True, + help='INPUT ONNX file path' + ) + parser.add_argument( + '-of', + '--output_onnx_file_path', + type=str, + required=True, + help='OUTPUT ONNX file path' + ) + parser.add_argument( + '-i', + '--input_names', + type=str, + action='append', + help='input names' + ) + parser.add_argument( + '-is', + '--input_shapes', + type=str, + nargs='+', + action='append', + help='input shapes' + ) + parser.add_argument( + '-o', + '--output_names', + type=str, + action='append', + help='output names' + ) + parser.add_argument( + '-os', + '--output_shapes', + type=str, + nargs='+', + action='append', + help='output shapes' + ) + + args = parser.parse_args() + INPUT_MODEL_PATH = args.input_onnx_file_path + OUTPUT_MODEL_PATH = args.output_onnx_file_path + INPUT_NAMES = args.input_names + INPUT_SHAPES = args.input_shapes + OUTPUT_NAMES = args.output_names + OUTPUT_SHAPES = args.output_shapes + + input_names = [name for name in INPUT_NAMES] + input_shapes = [shape for shape in INPUT_SHAPES] + output_names = [name for name in OUTPUT_NAMES] + output_shapes = [shape for shape in OUTPUT_SHAPES] + + input_dicts = {name:shape for (name, shape) in zip(input_names, input_shapes)} + output_dicts = {name:shape for (name, shape) in zip(output_names, output_shapes)} + + model = onnx.load(INPUT_MODEL_PATH) + updated_model = update_model_dims.update_inputs_outputs_dims( + model=model, + input_dims=input_dicts, + output_dims=output_dicts, + ) + + onnx.save(updated_model, OUTPUT_MODEL_PATH) \ No newline at end of file diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_nms_outputs_merge.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_nms_outputs_merge.py new file mode 100644 index 0000000000..296da32b96 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_nms_outputs_merge.py @@ -0,0 +1,64 @@ +#! /usr/bin/env python + +import torch +import torch.nn as nn +import onnx +from onnxsim import simplify +from argparse import ArgumentParser + + +class Model(nn.Module): + def __init__(self): + super(Model, self).__init__() + + def forward(self, batch, classid, x1y1x2y2, score): + batchno_classid_x1y1x2y2_score_cat = torch.cat([batch, classid, x1y1x2y2, score], dim=1) + return batchno_classid_x1y1x2y2_score_cat + + +if __name__ == "__main__": + parser = ArgumentParser() + parser.add_argument( + '-o', + '--opset', + type=int, + default=11, + help='onnx opset' + ) + args = parser.parse_args() + + model = Model() + + MODEL = f'24_nms_batchno_classid_x1y1x2y2_score_cat' + + onnx_file = f"{MODEL}.onnx" + OPSET=args.opset + + x1 = torch.ones([1, 1], dtype=torch.float32) + x2 = torch.ones([1, 1], dtype=torch.float32) + x3 = torch.ones([1, 4], dtype=torch.float32) + x4 = torch.ones([1, 1], dtype=torch.float32) + + torch.onnx.export( + model, + args=(x1,x2,x3,x4), + f=onnx_file, + opset_version=OPSET, + input_names=['cat_batch','cat_classid','cat_x1y1x2y2','cat_score'], + output_names=['batchno_classid_x1y1x2y2_score'], + dynamic_axes={ + 'cat_batch': {0: 'N'}, + 'cat_classid': {0: 'N'}, + 'cat_x1y1x2y2': {0: 'N'}, + 'cat_score': {0: 'N'}, + 'batchno_classid_x1y1x2y2_score': {0: 'N'}, + } + ) + model_onnx1 = onnx.load(onnx_file) + model_onnx1 = onnx.shape_inference.infer_shapes(model_onnx1) + onnx.save(model_onnx1, onnx_file) + + model_onnx2 = onnx.load(onnx_file) + model_simp, check = simplify(model_onnx2) + onnx.save(model_simp, onnx_file) + diff --git a/421_Gold-YOLO-Head/post_process_gen_tools/make_score_gather_nd.py b/421_Gold-YOLO-Head/post_process_gen_tools/make_score_gather_nd.py new file mode 100644 index 0000000000..c9cbc0f8a5 --- /dev/null +++ b/421_Gold-YOLO-Head/post_process_gen_tools/make_score_gather_nd.py @@ -0,0 +1,72 @@ +import os +os.environ['CUDA_VISIBLE_DEVICES'] = '-1' +import tensorflow as tf +import numpy as np +np.random.seed(0) +from argparse import ArgumentParser + + +if __name__ == "__main__": + parser = ArgumentParser() + parser.add_argument( + '-b', + '--batches', + type=int, + default=1, + help='batch size' + ) + parser.add_argument( + '-x', + '--boxes', + type=int, + default=5040, + help='boxes' + ) + parser.add_argument( + '-c', + '--classes', + type=int, + default=80, + help='classes' + ) + args = parser.parse_args() + BATCHES = args.batches + BOXES = args.boxes + CLASSES = args.classes + + + # Create a model + scores = tf.keras.layers.Input( + shape=[ + CLASSES, + BOXES, + ], + batch_size=BATCHES, + dtype=tf.float32, + ) + + selected_indices = tf.keras.layers.Input( + shape=[ + 3, + ], + dtype=tf.int64, + ) + + gathered_scores = tf.gather_nd( + scores, + selected_indices, + batch_dims=0, + ) + expands_scores = gathered_scores[:,np.newaxis] + + model = tf.keras.models.Model(inputs=[scores,selected_indices], outputs=[expands_scores]) + model.summary() + output_path = 'saved_model_postprocess' + tf.saved_model.save(model, output_path) + converter = tf.lite.TFLiteConverter.from_keras_model(model) + converter.target_spec.supported_ops = [ + tf.lite.OpsSet.TFLITE_BUILTINS, + tf.lite.OpsSet.SELECT_TF_OPS + ] + tflite_model = converter.convert() + open(f"{output_path}/nms_score_gather_nd.tflite", "wb").write(tflite_model) \ No newline at end of file diff --git a/421_Gold-YOLO-Head/url.txt b/421_Gold-YOLO-Head/url.txt new file mode 100644 index 0000000000..f3d10f3536 --- /dev/null +++ b/421_Gold-YOLO-Head/url.txt @@ -0,0 +1,5 @@ +https://github.com/huawei-noah/Efficient-Computing/tree/master/Detection/Gold-YOLO + +https://github.com/PINTO0309/onnx2tf +https://github.com/PINTO0309/simple-onnx-processing-tools + diff --git a/README.md b/README.md index f448ce4d83..df411697a4 100644 --- a/README.md +++ b/README.md @@ -217,6 +217,7 @@ I have been working on quantization of various models as a hobby, but I have ski |399|RetinaFace_MobileNetv2|[■■■](https://github.com/PINTO0309/PINTO_model_zoo/tree/main/399_RetinaFace_MobileNetv2)|||||||||||⚫|| |410|FaceMeshV2|[■■■](https://github.com/PINTO0309/PINTO_model_zoo/tree/main/410_FaceMeshV2)|⚫|⚫|⚫||⚫||⚫||||⚫|MediaPipe| |414|STAR|[■■■](https://github.com/PINTO0309/PINTO_model_zoo/tree/main/414_STAR)|⚫|⚫|⚫||⚫||⚫||||⚫|| +|421|Gold-YOLO-Head|[■■■](https://github.com/PINTO0309/PINTO_model_zoo/tree/main/421_Gold-YOLO-Head)|||||||||||⚫|Head (not Face)| ### 5. 2D/3D Hand Detection |No.|Model Name|Link|FP32|FP16|INT8|TPU|DQ|WQ|OV|CM|TFJS|TF-TRT|ONNX|Remarks| |:-|:-|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-|