Closed Wzj-zju closed 2 years ago
Hi, I cannot provide you the methods directly, since I don't work on the research position anymore and don't have the access to all the code, but I can clarify on the segmentation preprocessing pipeline. I believe parts of this are also described in the paper.
Essentially, Densepose segmentation model was used for body-part segmentations and paper from Liu et al. for clothing segmentations. Then, if neccessary, the body-part of clothing masks from two models were combined to get the final (provided) segmentations. I am not sure if that helps you, but these two models should suffice for generation of all segmentations used in the paper.
Can you please provide the preprocessing methods of image_body_parse and image_parse_with_hands.