Skip to content

Latest commit

 

History

History
343 lines (336 loc) · 29.6 KB

CHANGELOG.md

File metadata and controls

343 lines (336 loc) · 29.6 KB

1.6.0

  • 39b684f - (Javi Ribera) update README mentioning --ultrasmallnet
  • cb3f966 - (Javi Ribera) update version number and last modified dates
  • 7432e0b - (Javier Ribera) +option to make UNet smaller for pupil dataset
  • 02cb134 - (Javi Ribera) Update README (add link to CVPR Oral on Youtube)
  • f860f7c - (Javier Ribera) fix: input to the CNN could only be 256x256
  • 8316f72 - (Javi Ribera) update README: reduce image width
  • 8ff89fd - (Javi Ribera) +convergence gif in README
  • 5e0ed00 - (Javi Ribera) Add link to pre-trained models to README.txt
  • e6837b3 - (Javi Ribera) update CHANGELOG
  • dca0c98 - (Javi Ribera) fix: import future must go before copyright
  • dca0c98 - (Javier Ribera) fix: import future must go before copyright
  • e71f9d6 - (Javier Ribera) remove old unet_256x256_sorghum.ckpt checkpoint from git repo
  • e86144e - (Javier Ribera) README: explain code versioning
  • beeec39 - (Javier Ribera) fix README.md
  • 84b76dd - (Javier Ribera) add copyright notices and license
  • be6f1ac - (Javier Ribera) update author names
  • c78fc29 - (Javier Ribera) remove Dockerfile that is not maintained
  • b8c803e - (Javier Ribera) fix readme format
  • cc6fd12 - (Javier Ribera) Merge ../object-locator
  • a912ccb - (Javier Ribera) Create README.md
  • e1ba10c - (Javier Ribera) fix README file
  • bd6b9e0 - (Javier Ribera) added a new CHANGELOG.md
  • a6936a4 - (Javier Ribera) update version number in python package
  • ec0a64c - (Javier Ribera) update README
  • d5ae9d6 - (Javier Ribera) update README

1.5.0

  • d2ac92b - (Javier Ribera) + explanation in argparse
  • 1103d55 - (Javier Ribera) added tif as valid extension for imgs
  • 90122a4 - (Javier Ribera) random seed for initial split of validation dataset
  • dee532f - (Javier Ribera) automatically detect XML or CSV, and move plant stuff to separate module
  • fbafd71 - (Javier Ribera) new intermediate output w/ crosses on original img
  • 89da413 - (Javier Ribera) organize: move some dirs into intermediate dir
  • 6c4f31e - (Javier Ribera) fix: ignore nan thresholds when plotting kde
  • 49a079f - (Javier Ribera) split subrow_grid_location_yx into y and x
  • 0ca87b5 - (Javier Ribera) fix: saving empty CSV
  • 4fce7d1 - (Javier Ribera) fix: forgot to create output dir when not evaluating
  • 93265db - (Javier Ribera) fix: crash when there is not GT (just inference)
  • c0f7d07 - (Javier Ribera) use XML instead of CSV for inference
  • b7152bc - (Javier Ribera) fix: build dictionary with ints instead of tensors
  • 7536897 - (Javier Ribera) narrower code
  • d23931b - (Javier Ribera) update to pytorch 1.0.0
  • a3653ab - (Javier Ribera) fix: convert str to float when chekcing argparse
  • ba75223 - (Javier Ribera) force some arguments to be strictly positive
  • b20b568 - (Javier Ribera) argparse: dont show default if empty str
  • 4d5daaa - (Javier Ribera) narrower code
  • a87dcc3 - (Javier Ribera) mention which args are mandatory, and which optional
  • d5471fd - (Javier Ribera) update visdom to newer version
  • cbae028 - (Javier Ribera) move to XML, allow partial GT in XML, and 80-20 autosplit w/ XML
  • 91bd985 - (Javier Ribera) fix visdom not connecting w/ workaround
  • afcd771 - (Javier Ribera) wrap data loader generator in a function
  • 2b3ce85 - (Javier Ribera) make visdom optional, port modifiable, and print visdom hostname
  • 05f9531 - (Javier Ribera) fix (estimated map must be 1D)
  • 3fc0776 - (Javier Ribera) mention # images for training/validation
  • 1de232b - (Javier Ribera) go back to UNet
  • e590816 - (Javier Ribera) full EM
  • 9f2be09 - (Javier Ribera) new option to replace optimizer

1.4.0

  • e6d7e96 - (Javier Ribera) code cleanup

  • 691c47d - (Javier Ribera) removed non-used models

  • 4da6bb7 - (Javier Ribera) read from XML v0.4

  • 4e7ddeb - (Javier Ribera) fix: get_image_size in results script

  • 672187d - (Javier Ribera) max_ahd=diagonal in metrics_from_results script

  • c60c6e1 - (Javier Ribera) new script to make metrics CSV from results CSV

  • 4558009 - (Javier Ribera) trying CSRNet

  • 5b01aa7 - (Javier Ribera) merge BMM density and threshold in same plot

  • b9cb6c3 - (Javier Ribera) cap BMM plots at density=8 instead of 50

  • 09fa343 - (Javier Ribera) heatmap on top of original image during inference

  • 3fd4ca4 - (Javier Ribera) force R=1 when object count is known

  • dcb57d3 - (Javier Ribera) use default types of fixed # of objects

  • 918c438 - (Javier Ribera) Revert "use regressed pts in WHD loss"

  • 92f4a1a - (Javier Ribera) use regressed pts in WHD loss

  • 593e3ac - (Javier Ribera) nicer plots

  • 1ac6878 - (Javier Ribera) allow [] in taus in arguments

  • f7f44e4 - (Javier Ribera) move legend up in the plots

  • 783ab8d - (Javier Ribera) go back use SmoothL1Loss for regression

  • 87328ce - (Javier Ribera) new script to draw circles on top of images of a dataset

  • cca2313 - (Javier Ribera) remove unused variables

  • b103d7a - (Javier Ribera) fix: p=-1 by default to prevent NaNs

  • 2f0a63e - (Javier Ribera) fix: prevent NaNs by not using activ in last layers of U-Net

  • e4a8d2e - (Javier Ribera) fix: use seed in random transforms too

  • e369a3b - (Javier Ribera) paint red haircrosses on estimated pts during validation

  • 16f59ef - (Javier Ribera) go back to use Otsu thresholding

  • d42b959 - (Javier Ribera) use seed when shuffling dataset at the beginning

  • fe10e61 - (Javier Ribera) also paint white circles on top of training labels

  • 6d4361f - (Javier Ribera) paint red circles on top of estimated points and send to Visdom

  • b09b2c2 - (Javier Ribera) send heatmaps on top of original to visdom

  • 8d6c015 - (Javier Ribera) shuffle dataset at the beginning

  • 8c687dd - (Javier Ribera) fix: use mean of second Beta in BMM thresholding

  • 9f8642a - (Javier Ribera) use BMM instead of Otsu thresholding during validation

  • d4d1717 - (Javier Ribera) fix: tau is a float, not an int

  • 5cfec9c - (Javier Ribera) fix: skip validation every val_freq times

  • 9a1056b - (Javier Ribera) new script to find optimal learning rate (order of magnitude)

  • 226a5d6 - (Javier Ribera) new option --force to overwrite output testing results

  • c657a91 - (Javier Ribera) track and log running average loss

  • a30a792 - (Javier Ribera) also send avg loss to visdom

  • 12d7c09 - (Javier Ribera) fix: logger not accepting numbers

  • 7ab4177 - (Javier Ribera) fix: no need for model.train() for every batch (now faster)

  • 0f6dd56 - (Javier Ribera) added copyright info of BMM fitting

  • 012b527 - (Javier Ribera) nicer plots

  • 1a5a14a - (Javier Ribera) fix: multiple radii as input

  • df8a9d6 - (Javier Ribera) use Beta Mixture Model-based thresholding with tau=-2

  • d5b6c0d - (Javier Ribera) fit spherical GMM so it runs faster

  • e402b15 - (Javier Ribera) subsample mask points randomly so GMM fitting is faster

  • 7879ba4 - (Javier Ribera) fix: validation was not being done

  • 0a211d6 - (Javier Ribera) label otsu thresholding in metric plots

  • 8dca1e1 - (Javier Ribera) fix: take only some radii if too many to plot

  • 736ee04 - (Javier Ribera) rename variable windows->window_ids

  • 35369f7 - (Javier Ribera) fix: visdom not show validation because reusing window_ids

  • 0c77411 - (Javier Ribera) clearer print msg

  • 87a9228 - (Javier Ribera) fix: show always same # of decimals of tau

  • 4127738 - (Javier Ribera) fix: scale estimated map before saving as img

  • c5503e8 - (Javier Ribera) fix: multiple taus in argparse

  • dd6f056 - (Javier Ribera) use AMSgrad, the "convergence fix" for Adam

  • 6f941ea - (Javier Ribera) always printing losses should not be there besides for debugging

  • 2a57b29 - (Javier Ribera) rename softmin to generalized mean

  • 6a5efe2 - (Javier Ribera) p=9 by default

  • 0f86948 - (Javier Ribera) clean docstring

  • b15f18c - (Javier Ribera) new 2nd term in the cost function

  • 69afc86 - (Javier Ribera) clarify README instructions

  • 2aff47e - (Javier Ribera) no need to save checkpoint at the end of each epoch if no validation

  • c07194f - (Javier Ribera) smaller scatter markers

  • f28593c - (Javier Ribera) 50 taus instead of 100

  • f8f07f3 - (Javier Ribera) update last modified date

  • 605d9d3 - (Javier Ribera) added copyright notices to all files

  • 13a5d22 - (Javier Ribera) script is now part of the package:

  • 66e82f7 - (Javier Ribera) fix doc: setup.py said the package was inference-only

  • 788a995 - (Javier Ribera) able to send visualiz to remote Visdom server

  • 1e5b1ab - (Javier Ribera) conda environment: missing ballpark dependency

  • 7e3523f - (Javier Ribera) wrap clustering into a function

  • f709cb2 - (Javier Ribera) fix: undo accidental removal of line in commit d78ecf6

  • 07ff50b - (Javier Ribera) max AHD=max dist instead of inf during training

  • 257f37a - (Javier Ribera) encapsulate thresholding into a function

  • 82c59ee - (Javier Ribera) corner case: avoid crash when tau = -1 only

  • 7cfa32f - (Javier Ribera) better var names, torch0.4 optimizations, and use centroids_wrt_orig

  • 5144c82 - (Javier Ribera) mark Otsu result separate in the metric plots

  • a3a836e - (Javier Ribera) improve docstring

  • 7f7a26a - (Javier Ribera) fix: not reporting MAE

  • 639ba94 - (Javier Ribera) use item() from pythorch0.4

  • 905fce6 - (Javier Ribera) forgot Normalizer class

  • f873734 - (Javier Ribera) use Otsu thresholding in testing

  • efdf8fb - (Javier Ribera) show # params during testing

  • d5b5e31 - (Javier Ribera) normalize centroids in a new class

  • 9efadac - (Javier Ribera) show number of parameters

  • ca27de7 - (Javier Ribera) fix corner case when img has no object at all

  • dd18ba7 - (Javier Ribera) fix: forgot f of f-string

  • 875833b - (Javier Ribera) use non-default port number in visdom

  • 0b44f21 - (Javier Ribera) one painted output per threshold value

  • 664181b - (Javier Ribera) tau=1 => otsu thresholding

  • 83b963d - (Javier Ribera) change default visdom environment name

  • 3c641b1 - (Javier Ribera) also show r and R^2 on validation in visdom

  • f4b47e9 - (Javier Ribera) fix: dimension mismatch

  • e111d01 - (Javier Ribera) avoid matplotlib warning

  • 0460f85 - (Javier Ribera) corner case in WHD when there are no GT pts

  • 630dd4b - (Javier Ribera) shorter message

  • 9e659e2 - (Javier Ribera) use default types

  • d967556 - (Javier Ribera) seed is 0 now

  • 58eacaa - (Javier Ribera) also scale image during testing

  • 33bb408 - (Javier Ribera) pull info from CSV in a more explicit way

  • ef84a8d - (Javier Ribera) fix matplotlib warning by closing unused figures

  • d4ce8ee - (Javier Ribera) also compute r and R2 metrics

  • 1ef0014 - (Javier Ribera) plotting metrics inside py package, and plot when locating too

  • f0022c0 - (Javier Ribera) new script to plot multiple metrics from the results of a CSV

  • 2c1a828 - (Javier Ribera) use otsu thresholding

  • 81703fa - (Javier Ribera) fix CSV dataset loading

  • 23d441d - (Javier Ribera) fix

  • 2a9a589 - (Javier Ribera) fix: loading GT from CSV was broken because of other commits

  • be8a9f6 - (Javier Ribera) use CSV

  • d22a00e - (Javier Ribera) fix: loading GT from CSV was broken because of other commits

  • 657a039 - (Javier Ribera) fix: crash when spacing is not defined (provisional)

  • 44587d4 - (Javier Ribera) put additional info in output CSV

1.3.1

  • 25fd892 - (Javier Ribera) v1.3.1
  • dd1d718 - (Javier Ribera) no more wheels
  • 4bfbbf5 - (Javier Ribera) allow empty checkpoint directory
  • 8424c9a - (Javier Ribera) Update 'README.md'
  • e05370b - (Javier Ribera) Update 'README.md'
  • 376e145 - (Javier Ribera) show error when dataset directory is empty
  • 38a88b2 - (Javier Ribera) object-location -> object-locator
  • 0be6b72 - (Javier Ribera) Update 'README.md'

1.3.0

  • e8828f8 - (Javier Ribera) fix environment
  • 79328bd - (Javier Ribera) Update 'README.md'
  • 3f77343 - (Javier Ribera) Update 'README.md'
  • b27e4c2 - (Javier Ribera) update README
  • 82db2db - (Javier Ribera) environment.yml for Windows
  • 127d1e3 - (Javier Ribera) v1.3.0
  • a24827a - (Javier Ribera) change deprecated class to remove warning
  • 82f62f5 - (Javier Ribera) better argparse help
  • 01b06ab - (Javier Ribera) update README

1.2.0

  • 6a1253a - (Javier Ribera) v1.2.0
  • fb9f2e0 - (Javier Ribera) train on GPU, test on CPU
  • 019de1e - (Javier Ribera) update README
  • 39320e1 - (Javier Ribera) no more wheel files
  • d89a4ce - (Javier Ribera) add copyright notice
  • 18a6fa8 - (Javier Ribera) fix: cornercase break when estimation point set is empty
  • aea9ad7 - (Javier Ribera) fix: not loading default checkpoint
  • f8df58c - (Javier Ribera) forgot old model ckpt
  • 1165e2c - (Javier Ribera) fix: fail testing whithout --evaluate
  • ca755b9 - (Javier Ribera) Update 'README.md'
  • 8e16c56 - (Javier Ribera) clean environment file
  • f59768e - (Javier Ribera) allow XMLv0.3
  • 85d94d8 - (Javier Ribera) compute stats in cm instead of px
  • 84e70db - (Javier Ribera) rename script
  • 36e43c4 - (Javier Ribera) use argparse for csv stats
  • db3e9e5 - (Javier Ribera) allow test on CPU
  • a8c19cd - (Javier Ribera) use peterpy to show status
  • 43b2f47 - (Javier Ribera) update to pytorch v0.4
  • cc428fa - (Javier Ribera) fix: adapt training to new Judge class
  • 44404de - (Javier Ribera) fix: cornercase when prec=0 and rec=0
  • aee0224 - (Javier Ribera) more flexible logger
  • 5f68638 - (Javier Ribera) fix: cornercase when len(c)==0
  • 2ae0b56 - (Javier Ribera) fix: saving and resuming from same checkpoint
  • 1738d20 - (Javier Ribera) add option to evaluate metrics or not
  • f260657 - (Javier Ribera) move metrics inside Judge so locate.py is cleaner
  • dc40e52 - (Javier Ribera) fix: crash when testing without GT
  • e6bbf2f - (Javier Ribera) measure fscore besides precision n recall
  • 00d824f - (Javier Ribera) also print Mean (Percent) Error w/o Absolute (to check bias)
  • db9adb2 - (Javier Ribera) indicate r in the output CSV of prec and rec
  • c279c1c - (Javier Ribera) script to make plot now accepts CSV and uses argparse
  • c0463c5 - (Javier Ribera) rename script to make precision and recall plot
  • ec94d50 - (Javier Ribera) also export locations in CSV
  • 1ed5af7 - (Javier Ribera) autoformat
  • 0810141 - (Javier Ribera) fix: paint by default
  • 55dffe4 - (Javier Ribera) save and compute metrics using original image size
  • 2ed82ff - (Javier Ribera) use size before resize to normalize Euc distance in WHD
  • c23d015 - (Javier Ribera) fix: validation dataset not being resized
  • b1f4caa - (Javier Ribera) dataset can be in XML, and resize input image on the fly
  • 598f1ff - (Javier Ribera) fix: when input images are not rectangular
  • ee5ce8c - (Javier Ribera) bash script to build a wheel file for pip
  • 24b8cee - (Javier Ribera) fix: crash when save path was not provided
  • 2145d31 - (Javier Ribera) fix README
  • e92d5f6 - (Javier Ribera) +conda instructions
  • 2623eaf - (Javier Ribera) new wheel
  • 60a5172 - (Javier Ribera) no need for main
  • f945bde - (Javier Ribera) get rid of the plant word
  • 68f5d9c - (Javier Ribera) rename module name to get rid of plants
  • db5e30e - (Javier Ribera) make train.py callable as module
  • 78578d2 - (Javier Ribera) fix: epoch+=1
  • ae2492f - (Javier Ribera) tunable LR for Adam
  • a6f99fd - (Javier Ribera) fix: dont save at end of epoch if not asked
  • da8e7a1 - (Javier Ribera) +tqdm bar w/ running avg of loss in an epoch
  • cd05fb4 - (Javier Ribera) option to drop last batch
  • 076345d - (Javier Ribera) option to train w/ Adam vs SGD
  • 6600cab - (Javier Ribera) ignore val set when val-freq=0
  • 9192425 - (Javier Ribera) if no val set, store checkpoint at each end of epoch
  • a35696f - (Javier Ribera) separate file for argparse
  • 10d819d - (Javier Ribera) threshold value we always used
  • bdbdb0a - (Javier Ribera) also evaluate MAE, RMSE in validation
  • 805acac - (Javier Ribera) testing script also evaluates MAE, MSE, RMSE
  • 2145f9f - (Javier Ribera) testing script also saves CSV w/ precision n recall
  • dd30bd1 - (Javier Ribera) fix: old var name
  • eab3cd6 - (Javier Ribera) fix: known number of points with batches
  • addc59c - (Javier Ribera) also test recall
  • 45a4729 - (Javier Ribera) autoformat
  • 90e5048 - (Javier Ribera) new f-string
  • fea1016 - (Javier Ribera) fix: tensor dimensionality changed because bs>1
  • 0101822 - (Javier Ribera) show r5-precision/recall in visdom
  • d6a9b15 - (Javier Ribera) indicate when len(mask)==0
  • d5264bc - (Javier Ribera) fix: validation frequency
  • 885a986 - (Javier Ribera) show epoch instead of iteration # in visdom
  • 66e2e54 - (Javier Ribera) validate every F epochs
  • 33a1031 - (Javier Ribera) SGD->Adam
  • f2fc7f7 - (Javier Ribera) smooth L1 for regression, also plot MAPE
  • ae1e364 - (Javier Ribera) force batch size to 1 during evaluation
  • 18e3cf2 - (Javier Ribera) indicate lambda in visdom
  • 6d4e79e - (Javier Ribera) wrong default
  • bab91cc - (Javier Ribera) smooth L1 regression and avoid /0
  • 561311d - (Javier Ribera) update locate (test)
  • a860c77 - (Javier Ribera) both inner and outter regression
  • 44ae265 - (Javier Ribera) code cleanup
  • 018cabf - (Javier Ribera) test was renamed to locate
  • ace9e47 - (Javier Ribera) Variable is needed by logger
  • 8cafc6f - (Javier Ribera) code cleanup
  • 27a181e - (Javier Ribera) no more references to plants
  • 6d07d32 - (Javier Ribera) not using this
  • bb749d5 - (Javier Ribera) volatile Variables during inference
  • e803292 - (Javier Ribera) option to disable data augmentation
  • 6c7e2e5 - (Javier Ribera) move data loading code to data.py
  • d61c9ed - (Javier Ribera) allow batch size > 1
  • a13c7db - (Javier Ribera) modularize visdom logger code
  • 4531d92 - (Javier Ribera) update to pythorch 0.3 and fix 1x1 normalization layer
  • a3a0c1c - (Javier Ribera) (%) in prec and recall in visdom
  • 76648df - (Javier Ribera) tensortype and update visdom
  • 6594334 - (Javier Ribera) rename file
  • 8e01e9b - (Javier Ribera) organized files
  • 28cc581 - (Javier Ribera) delete deliverable old code already moved
  • f4e7fd1 - (Javier Ribera) merge deliverable w/ new code. test may break
  • 5e055cb - (Javier Ribera) rename some variables
  • b8fe6dc - (Javier Ribera) move data classes to new file and do RandomFlip
  • e4a65a4 - (Javier Ribera) go back to 256x256 inputs
  • fb0ae50 - (Javier Ribera) fix environment file w/ unfindable packages
  • 9b8a7d1 - (Javier Ribera) installation instructions for docker
  • fb5ff54 - (Javier Ribera) remove prefix from conda environment
  • ede2e98 - (Javier Ribera) ascii only
  • 2a3bf78 - (Javier Ribera) always allow printing unicode characters
  • 8b2c44e - (Javier Ribera) Dockerfile
  • b06c958 - (Javier Ribera) update wheel
  • 9d2554d - (Javier Ribera) remove unncesessary dependencies
  • 52c2e10 - (Javier Ribera) fix warning: if call from bash, use main
  • 370955c - (Javier Ribera) include model checkpoints inside package wheel
  • 66f9edc - (Javier Ribera) prepare wheel of plant-locator
  • 733ddf3 - (Javier Ribera) organize for wheel deliverable
  • 42bdf4d - (Javier Ribera) CSV is now optional if you dont have GT on testset
  • 6254910 - (Javier Ribera) fix warning: rename deprecated class
  • 0071240 - (Javier Ribera) no need to print architecture for inference
  • 947a8a8 - (Javier Ribera) rename and code cleanup
  • a30e9c5 - (Javier Ribera) more flexible input image size, but forced by checkpoint
  • 79a7646 - (Javier Ribera) name checkpoint -> model
  • 477b4dc - (Javier Ribera) trying to make CUDA optional, but cant for now => mandatory
  • 2c7070b - (Javier Ribera) fix backward compatibility with old chkpts
  • 800a155 - (Javier Ribera) who cares about epoch during testing
  • 58f88ce - (Javier Ribera) gitignore
  • ed3457d - (Javier Ribera) go back to 256x256 images
  • a544881 - (Javier Ribera) code cleanup
  • c66cb5d - (Javier Ribera) clean argparse descriptions
  • 674c40d - (Javier Ribera) clean code + argparse description
  • 8090356 - (Javier Ribera) prepare inference-only deliverable
  • 9aa22c9 - (Javier Ribera) generate_csv.py needs cv2 to get image size
  • 2539326 - (Javier Ribera) allow fix number of estimated points
  • 4fe1121 - (Javier Ribera) update csv parser to new XML format
  • 31c714d - (Javier Ribera) new script to analyze plant spacing (histograms of means, stds...)
  • f2f14cd - (Javier Ribera) script to generate CSV from Yuhao
  • adf5100 - (Javier Ribera) accumulate precision and recall for multiple r
  • 4b18016 - (David Güera) Script to generate result figures (uses xlsx)
  • 2b096eb - (Yuhao Chen) put the file with all dataset directorys, i will loop through all of them except 12 (which doesn't exist)
  • 0e41a05 - (Javier Ribera) Im breaking my own rule and commiting a lot of crap together :)
  • 8cb1ff3 - (Yuhao Chen) added precision and recall evaluating function
  • 29bdd8c - (David) Includes modification to estimate plants without regressor model
  • b52d635 - (Yuhao Chen) added result parsing script
  • 095b3a1 - (Javier Ribera) fixes on default arguments
  • 6df1781 - (Javier Ribera) removed useless parameters
  • e1fc339 - (Javier Ribera) fix badly replaced files
  • e42f6db - (Javier Ribera) conda environment
  • fcb2984 - (Javier Ribera) init commit, very advanced status