keras-cv
keras-cv copied to clipboard
Vectorize CropAndResize
#1415 cc @LukeWood @ianstenbit
Results
@soma2000-lang looks like the unit tests are failing!
@LukeWood looking into this.Actually keras-cv is not properly installed in my system because after some system update the pip package did not seem to work.I am trying to fix the problem with pip right now.
@LukeWood I am trying to fix the pip issue with my system.Till I fix this,
I just run the test using Google Colab and it seems to pass.
Link -https://colab.research.google.com/drive/1-TZpbQDMp1fe3mrfvouWAqH_PISf9SBh?usp=sharing
it will be great if you check it once.
@soma2000-lang left a comment including a merge conflict. Can you resolve?
@LukeWood Resolved
@soma2000-lang the goal is to independently augment each image. I think getting this to work will require you to update the logic a bit.
@LukeWood I made required changes.But some tests are still failing!
It looks like this is a legitimate failure - can you fix the test failures? @soma2000-lang
@LukeWood I sent you an email 2 days back .The emIl id starts with seckroll16 .I would be really greatfull if you can check out once.😊
Regards
@james77777778 It would be great if youvcan help me with this.I am stuck on this for long.
Hi @soma2000-lang
I have roughly read your PR and I can share the idea about self.get_random_transformation_batch
https://colab.research.google.com/drive/15ed5Fqn5zrQClmmuvB1DWrDGwMrHA4Ka?usp=sharing
I think the vectorizing could be complete by modifying self._transform_bounding_boxes and self._crop_and_resize based on batched and independently sampled transformations
@james77777778 Thanks!
@soma2000-lang can you add a test like the one in https://github.com/keras-team/keras-cv/pull/1480/files#diff-1c771135874332a25b437da4a5706faea85ff719331dfe7beb255403a1a66477
@LukeWood sure!.Also sorry for the delay I will try to finish this pr soon.
@soma2000-lang can you please rebase with master?
Sure