search

liming-ai / ControlNet_Plus_Plus

Official PyTorch implementation of ECCV 2024 Paper: ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback.
https://liming-ai.github.io/ControlNet_Plus_Plus
Apache License 2.0
380 stars 16 forks source link

About normalize #12

Open Strike1999 opened 2 weeks ago

Strike1999 commented 2 weeks ago

Thanks for your work!

For image segmentation, you take image = normalize(image, (0.485, 0.456, 0.406), (0.229, 0.224, 0.225)). However, when inferencing reward model outputs = reward_model(image.to(accelerator.device)), does this step inherently include image normalization as part of the MMsegmentation test pipeline?

Thanks again.

liming-ai commented 2 weeks ago

Thanks for your work!

For image segmentation, you take image = normalize(image, (0.485, 0.456, 0.406), (0.229, 0.224, 0.225)). However, when inferencing reward model outputs = reward_model(image.to(accelerator.device)), does this step inherently include image normalization as part of the MMsegmentation test pipeline?

Thanks again.

Thank you for your attention. We call the seg model forward function that defined in MMSeg, instead of directly calling the entire MMSeg Pipeline for inference, so it will not be affected by normalization.