Allow VideoProcessor to Accept Single Image Inputs#13084
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
|
||
| def postprocess_video( | ||
| self, video: torch.Tensor, output_type: str = "np" | ||
| self, video: torch.Tensor, output_type: str = "np", **kwargs |
There was a problem hiding this comment.
Currently, this would facilitate passing the do_denormalize flag to VaeImageProcessor.postprocess. But it's intended more as a forward-looking change which allows postprocess_video to support any arguments that postprocess might want.
sayakpaul
left a comment
sayakpaul
left a comment
There was a problem hiding this comment.
Just one comment. I guess we can create an LTX-2 specific video processor subclassing from the current one and implement center cropping logic?
|
For LTX-2, the idea is that after the changes in this PR, we can call 
|
|
@yiyixuxu WDYT? |
3 participants
What does this PR do?
This PR adds support for single-image inputs to
VideoProcessor.preprocess_video. AsVideoProcessor.preprocess_videousesVaeImageProcessor.preprocessunder the hood, the PR also changespreprocess_videoto forward keyword arguments topreprocess. This allows it to support arguments thatpreprocesssupports, such asresize_mode.Changelist
videos forVideoProcessor.preprocess_video.VideoProcessor.preprocess_videoandVideoProcessor.postprocess_videonow forward keyword arguments toVaeImageProcessor.preprocessandVaeImageProcessor.postprocess, respectively.VideoProcessor.Inspired by discussion in #13058.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.
@sayakpaul
@yiyixuxu