General question

[antithing]

Hi, and thanks for making this code available!

General question, how are the segments chosen by the model?

My use case is human segmentation with a complex background.

Which of your models would be most suitable? Would this work directly, or would passing a box be better?

Thanks!

[ZhengPeng7]

The portrait or general should be the best ones. You can use my HF demo to make a comparison: https://huggingface.co/spaces/ZhengPeng7/BiRefNet_demo.

If the scene is very complex and the human might not be the most dominant/salient subject, passing a box may be better to eliminate the distractors.