keras-cv
keras-cv copied to clipboard
keras-team
Update current FPN for the extra level support. This is an extension that is used by the Retina-net application
What does this PR do?
Update current FPN for the extra level support. This is an extension that is used by the Retina-net application.
We could use this as the proper replacement for https://github.com/keras-team/keras-cv/blob/master/keras_cv/models/object_detection/retina_net/internal/layers/feature_pyramid.py
Fixes # (issue)
Before submitting
- [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
- [x] Did you read the contributor guideline, Pull Request section?
- [ ] Was this discussed/approved via a Github issue? Please add a link to it if that's the case.
- [x] Did you write any new necessary tests?
- [ ] If this adds a new model, can you run a few training steps on TPU in Colab to ensure that no XLA incompatible OP are used?
Chatted with @tanzhenyu offline for this. We will park this PR for now since this is only going to be used by the Retina-net and not blocking the segmentation model. We will revisit this when we clean up the Retina net model code.
Chatted with @tanzhenyu offline for this. We will park this PR for now since this is only going to be used by the Retina-net and not blocking the segmentation model. We will revisit this when we clean up the Retina net model code.
Sounds good to me.
Chatted with @tanzhenyu offline for this. We will park this PR for now since this is only going to be used by the Retina-net and not blocking the segmentation model. We will revisit this when we clean up the Retina net model code.
Have you chatted about RetinaMask for the segmentation to reuse the same RetinaNet backbones? https://arxiv.org/abs/1901.03353
Have you chatted about RetinaMask for the segmentation to reuse the same RetinaNet backbones?
See also the related thread we had with @tanzhenyu at https://github.com/keras-team/keras-cv/issues/920#issuecomment-1278279734
Thanks for the work on this @qlzh727 - @jbischof has been working on a new version of the FPN, and I think that the design has changed a LOT since this PR - so lets close this for now. Please re-open if you feel there is something in here we should consider, and we can update the code going forwards!