@rohitmidha23 and me have been following this challenge for quite a while. We have written a starter notebook using MaskRCNN. We further integrate MaskRCNN with WandB which really helps to keep track of the various experiments that anyone might want to run. Check out our runs here
We also added functionality to view predictions of the model on the validation set as the model trains.
You can find the notebook @ https://colab.research.google.com/drive/1D8jC9GdHhCyoGB-8bJogW-lSO21MeTKF?usp=sharing
We trained the model on 40 classes alone due to Colab’s timing restrictions. However we provide a GitHub repo using which you can train on the entire dataset.
We also made a submission repo, using which you can submit models trained using the code above. The model we trained gets a precision score of 0.135 on the test set.
We hope that this notebook helps out other participants.
As always we are open to any feedback, suggestions and criticism!
If you found our work helpful, do drop us a !