Port Faster R-CNN to Keras3

[sineeli]

What does this PR do?

Port legacy Faster R-CNN to Keras 3

Fix minor bug in ROI Align, make sure to build the shapes for each layer components.

The present PR is modified code of Updating FasterRCNN to use Task API contributed by @ariG23498. Please @ariG23498 any suggestions form your side are appreciated.

[sineeli]

Can we make last dimension of the roi_generator.py boxes output have fixed 4 as dimension, any suggestion on how to make it possible ?

Ref:

[sineeli]

@VarunS1997 Review for any corrections are fixes required from based code.

Thanks!

[ariG23498]

Hey @sineeli this is so cool! Thanks for taking this up. It was a long time ago and unfortunately I would not be able to give you a thorough review.

What I understand should be green flags are:

• The model should follow the Functional API of Keras
• It should use the model.fit() API and train on PASCAL VOC

Last I worked on it I could not get every component to comply with the Functional API. If you got that part sorted, you are good to go ^_^

Closes #2012

[sineeli]

Hey @sineeli this is so cool! Thanks for taking this up. It was a long time ago and unfortunately I would not be able to give you a thorough review.

What I understand should be green flags are:

• The model should follow the Functional API of Keras
• It should use the model.fit() API and train on PASCAL VOC

Last I worked on it I could not get every component to comply with the Functional API. If you got that part sorted, you are good to go ^_^

Closes #2012

Yes trying to make it functional, hopefully it will workout!

[VarunS1997]

Generally, good ideas here and good implementation.

We should definitely, however, implement some of the requested changes.

Also please remember to run keras_cv's linter included under the shell/ directory.

I also will have to come back to this for architecture verification. But, if you have a working colab or similar, that would work as well.

[divyashreepathihalli]

left some comments to reduce tests sizes.

[divyashreepathihalli]

Excellent work!! Thanks Siva!!

FasterRCNN has been handed off to Divya.