search

keras-team / keras-cv

Industry-strength Computer Vision workflows with Keras
Other
1.01k stars 331 forks source link

How does Keras-YOLOv7 deal with conv2d layer padding bug? #2379

Closed abhaskumarsinha closed 7 months ago

abhaskumarsinha commented 8 months ago

Current Behavior:

Mismatch of the elements in the edges and corners in Keras Conv2D Layer when comparing the conv2d of YOLOv9 that was implemented on PyTorch.

Expected Behavior:

No mismatch.

Steps To Reproduce:

Version:

Anything else:

I believe the code for YOLOv9 was extended from YOLOv7 that has already been implemented in KerasCV. How did they not encounter this issue? Is there any problem on my side?

Linked: https://github.com/keras-team/keras/issues/19291#issue-2180964867

abhaskumarsinha commented 8 months ago

I just noticed Keras-CV supports only v8 and no support for v7 now...

innat commented 8 months ago

YOLO-V7 https://github.com/keras-team/keras-cv/issues/1179

abhaskumarsinha commented 8 months ago

@innat I see, the padding of the element in the edges are manually implemented here.

abhaskumarsinha commented 8 months ago

@innat Ah, the problem with that implementation is that I won't get to use to switch to backends after using that... If I switch from the TensorFlow backend to the PyTorch backend using that code, the code would break. There's no point in implementing that if that doesn't get fixed! If that doesn't remain backend invariant, there's no point of using Keras at all then!

Let me post the code to the main issue so that gets attention.

github-actions[bot] commented 8 months ago

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

github-actions[bot] commented 7 months ago

This issue was closed because it has been inactive for 28 days. Please reopen if you'd like to work on this further.