Feature linux support

[CaptainDario]

This PR should add linux support.

Environment:

• Ubuntu 22.04

Changes:

• Load TF Lite binary
• update examples (same as macos) to support linux

Notes:

• Camera is disabled for the examples on desktop as this is not really supported

Progress:

• load TF Lite on linux
• test examples
  • audio classification
    • uses some platform channels that do not work on linux
  • image classification mobilenet
    • only from file works
  • live object detection ssd mobile net
    • uses camera package that only works on iOS / mobile
  • object detection ssd mobile net
    • only from file works
  • style transfer
    • works
  • super resolution esrgan
    • works
  • text classification
    • works

[ghost]

@st-duymai Do you happen to have a Linux machine? I don't think I'll be back at the office before leave for doing this review. If not, I can ask someone internally, too. Thanks!

I don't currently have a Linux machine, but I can set one up to test it.

@CaptainDario Could you help me resolve the conflict?

[CaptainDario]

Yeah, sure! Do you encounter the same issue as PaulTR?

[CaptainDario]

@PaulTR I think I found the problem that you were encountering and I pushed a fix. @st-duymai Let me know if you have problems setting this up.

[PaulTR]

Sounds good, I'll see if I can make my way in to an office tomorrow or Wednesday to give it one more shot. Thanks!

[ghost]

@CaptainDario I frequently encounter this problem when building Linux applications. Have you ever dealt with this issue before?
/home/ubuntu/Documents/flutter-tflite/example/super_resolution_esrgan/build/linux/x64/debug/bundle/super_resolution_esrgan: symbol lookup error: /home/ubuntu/Documents/flutter-tflite/example/super_resolution_esrgan/build/linux/x64/debug/bundle/blobs/libtensorflowlite_c-linux.so: undefined symbol: _ZN6tflite3ops7builtin12Register_NEGEv

[CaptainDario]

No I have not seen this issue, how did you build the binary?

The one I used can be found here

[ghost]

Thanks @CaptainDario I will check again with your .so library. I may have built it wrong at some step.

[ghost]

Hi @PaulTR ,

After check with .so library that @CaptainDario provide above, I can confirm that the following five examples are working properly:

1. Style transfer
2. Text classification
3. Super resolution esrgan
4. Object detection SSD Mobilenet(only file work)
5. Image classification

However, two examples are not working:

1. Audio classification
2. Live object detection SSD Mobilenet

[ghost]

Hi @CaptainDario
Here is my libtensorflow_cc.so build step. Is this step correct?

sudo apt-get install cmake
git clone https://github.com/tensorflow/tensorflow.git tensorflow_src
mkdir tflite_build
cd tflite_build
cmake ../tensorflow_src/tensorflow/lite/c
cmake --build . -j

[CaptainDario]

Hi @CaptainDario Here is my libtensorflow_cc.so build step. Is this step correct?

sudo apt-get install cmake
git clone https://github.com/tensorflow/tensorflow.git tensorflow_src
mkdir tflite_build
cd tflite_build
cmake ../tensorflow_src/tensorflow/lite/c
cmake --build . -j

You forgot to checkout a specific release, I guess the state you are currently in is broken. I used r2.12

[PaulTR]

Hi @PaulTR ,

After check with .so library that @CaptainDario provide above, I can confirm that the following five examples are working properly:

1. Style transfer
2. Text classification
3. Super resolution esrgan
4. Object detection SSD Mobilenet(only file work)
5. Image classification

However, two examples are not working:

1. Audio classification
2. Live object detection SSD Mobilenet

Awesome, thanks for checking on those. Yeah those two are expected to not work with desktop, so that's fine. I think once we have those conflicts resolved we should be pretty golden then for all three desktop versions! :)

[PaulTR]

Just confirmed image classification on my work machine with my own .so file, so we're looking good on my side too

[ghost]

Hi @CaptainDario Would you mind resolving the conflict in this PR? Thank you.

[CaptainDario]

Yeah sure! I did not notice that there are conflicts.

[CaptainDario]

done

[ghost]

Hi @CaptainDario . I think this PR look greate and ready to merge. I just have one minor request regarding the readme file, which I mentioned earlier. Thank you for your hard work.

[ghost]

LGTM. Thanks

[CaptainDario]

@st-duymai are you able to release new tf lite version on pub?

[ghost]

@st-duymai are you able to release new tf lite version on pub?

Sorry. I don't have any information about that.

[CaptainDario]

@PaulTR, when you have some time, could you take a look at releasing a new version at pub?

[ghost]

@CaptainDario The new version was published a few days ago (tflite_flutter: ^0.10.4). Please check.

[CaptainDario]

@st-duymai thanks for the heads up, I tried it!