Auto annotation GPU support

[jahaniam]

Motivation and context

• Being able to run a model on GPU is one of the main advantages of using deep learning based auto annotation tool.
  GPU support was added on nuclio version 1.5.0 on this PR (although they haven't updated their documentation yet). By adding GPU resources to 1 or any positive number, the flag --gpus=all will be added during the container creation.

• Documentation was wrong. It was telling to download the latest nuclio while the dashboard on docker-compose was running on nuclio version 1.4.8. This will result in missmatch and some build issues depending on the versions.

• Fixed a bug in fast-rcnn-tensorflow code when the image is png or grayscale (if the image has more or less than 3 channels). This bug might exist in some of the other models as well. I might do more PRs in the future.

• This PR is a sample model opening the path for future models as we go users can add more models here.

• Still we can send images in batches for future optimizations (not implemented)

• This will also cover Nuclio Automatic/SemiAutomatic AI Tool Functions not running on GPU #2489 Serverless component nuclio doesn't support nvidia-docker / GPU acceleration. #1997 improved documentation for component 'Semi-automatic and automatic an… #2273 Could not get models from the server #2541 and some more.

How has this been tested?

by checking docker logs and docker exec into the container and making sure nvidia-smi works and code runs on the GPU.

Checklist

• I submit my changes into the develop branch
• I have added description of my changes into CHANGELOG file
• I have updated the documentation accordingly
• I have added tests to cover my changes
• I have linked related issues (read github docs)
• I have increased versions of npm packages if it is necessary (cvat-canvas,
  cvat-core, cvat-data and cvat-ui)

License

• I submit my code changes under the same MIT License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below)

# Copyright (C) 2020 Intel Corporation
#
# SPDX-License-Identifier: MIT

[coveralls]

Coverage increased (+0.3%) to 62.131% when pulling 18c18dd on jahaniam:develop into 76ff7ef on openvinotoolkit:develop.

[azhavoro]

@jahaniam Thank you for the contribution, I'll test PR today.

[jahaniam]

@jahaniam Thank you for the contribution, I'll test PR today.

Looking forward to more collaborations.

I will slowly add the GPU support for other models as well. Next in line is MaskRCNN or adding support for detection2 framework(by facebook).

[nmanovic]

@jahaniam , awesome contribution. I will look at the PR as well. Give us sometime. Really want to add your changes before our next release.

[jahaniam]

@nmanovic I searched a lot but I couldn't find the exact place where the response of the nuclio functions are getting processed. For example, for this fast-rcnn, we send an image to the nuclio function and we get the labels and bounding box (rectangle type with points) as a response. But where this information is being processed in cvat? I appreciate if you can pinpoint me to line of the code that is processing the response from nuclio (or sending the request to nuclio). This will help me a lot in bug fixing and further contributing to the CVAT.
I could track it to the cvat rest API but no luck. Thank you.

[nmanovic]

@nmanovic I searched a lot but I couldn't find the exact place where the response of the nuclio functions are getting processed. For example, for this fast-rcnn, we send an image to the nuclio function and we get the labels and bounding box (rectangle type with points) as a response. But where this information is being processed in cvat? I appreciate if you can pinpoint me to line of the code that is processing the response from nuclio (or sending the request to nuclio). This will help me a lot in bug fixing and further contributing to the CVAT.
I could track it to the cvat rest API but no luck. Thank you.

https://github.com/openvinotoolkit/cvat/tree/develop/cvat/apps/lambda_manager

[jahaniam]

All comments have been addressed accordingly. I appreciate it if you can review it again. @nmanovic @azhavoro

[nmanovic]

@jahaniam , great contribution! Thank you very much for you time. 👍

[jahaniam]

@jahaniam , great contribution! Thank you very much for you time.

Thank you for open-sourcing this awesome software and being so responsive.

[jahaniam]

Can you reopen the pr please?
sorry, I forgot to update the changelogs. I have updated that in my forked repo. Should I do another pr for that?

[jahaniam]

@nmanovic