[Help wanted] FCOS: Fully Convolutional One-Stage Object Detection

[jaeyounkim]

Help wanted: Research paper code and models

Please help us to implement the following research paper using TensorFlow 2.

If you want to contribute, please leave a comment to express your interest.

Awesome external contributors will be nominated for Google Open Source Peer Bonus.

Research papers

• FCOS: Fully Convolutional One-Stage Object Detection

Status

• In progress

Contributor(s)

• @paxbun

Tasks

• Implement a paper in TensorFlow 2
• Reproduce the paper results
• Publish documentation
• Publish pre-trained SavedModel files at TensorFlow Hub

Development environment

• Branch: https://github.com/tensorflow/models/tree/fcos

Requirements

• Follow our contribution guidelines

License

By contributing, you agree that your contributions will be licensed under Apache License 2.0.

[raikonenfnu]

Hey @jaeyounkim, my friends @nithinsubbiah, @stwinata and I are interested in contributing and implementing this paper. It's our first time contributing to tensorflow/models (typically we play around MLIR land HLO/Linalg for nextgen tf compilers), Would the project be structured in this manner? Thanks in advance!

[paxbun]

Hi @raikonenfnu, I'm already working on this paper (I've just finished reading this paper now and this is the first time for me too!), so we need some rearrangements if you and your friends want to implement this paper.

[raikonenfnu]

hi @paxbun, thanks for letting me know. that's alright, we can look for other papers to implement! Goodluck on this implementation, hope you have a lot of fun! :)

[Tanishq-01]

Is this still open @jaeyounkim . I would be happy to add if open

[paxbun]

@Tanishq-01 I have failed to complete the implementation; for some reason, it successfully converges on Google Colab but not on my GPU server with A100. As far as I know, I'm no longer participating in the program. I apologize for all the extra costs for waiting for me.

[Samir-atra]

Hello,

I have started reading this paper and will start working on implementing it, it is my first time fully implementing a paper from the beginning to the end. and I wanted to let you know.

Kind regards
Samer

[Samir-atra]

Hello,
@jaeyounkim @paxbun
I have finished part of the implementation, and the model compiles today. There are some fine edits to the architecture that need to be made, and then I start training, and today, I do not have access to the necessary resources to run the model and load the COCO dataset. If you can help me with access to computation resources, I will be able to move forward.

I will be looking forward to hearing from you.

Kind regards
Samer Attrah

[Samir-atra]

Hello,

I just created a pull request,
Please note that it is a request for computation resources sponsorship and an initial implementation that I was able to run and test on Kaggle on a tpu environment and CPU only, with a minimal number of images for training included as a dataset.

Original implementation at: FCOS_Implementation

I look forward to hearing from you about the computation resources sponsorships. Please help

Kind regards
Samer