https://github.com/dvlab-research/ControlNeXt/blob/ab4b3acf912cc178d23bbf003369dfb657fc8d01/ControlNeXt-SVD-v2-Training/utils/vid_dataset.py#L167
Hi, nice work! Why is the BGR2RGB transformation applied only to the guide_values (and not to the reference images)?
Additionally, the VideoReader records RGB output, so I believe this transformation (BGR2RGB)might distort the original images. Even though it may not cause significant damage to the skeleton images (guide_values), there doesn’t seem to be a similar transformation during the reference stage.
https://github.com/dvlab-research/ControlNeXt/blob/ab4b3acf912cc178d23bbf003369dfb657fc8d01/ControlNeXt-SVD-v2-Training/utils/vid_dataset.py#L167
Hi, nice work! Why is the BGR2RGB transformation applied only to the guide_values (and not to the reference images)?
Additionally, the VideoReader records RGB output, so I believe this transformation (BGR2RGB)might distort the original images. Even though it may not cause significant damage to the skeleton images (guide_values), there doesn’t seem to be a similar transformation during the reference stage.