Use cloth segmentation pretrained model both for training and inference. For training, it's a convenient augmentation strategy, similar to what Sengupta's proposes in "Synthetic Training..." paper. For inference, of course, we somehow need to estimate the segmentation maps.
Description
Use cloth segmentation pretrained model both for training and inference. For training, it's a convenient augmentation strategy, similar to what Sengupta's proposes in "Synthetic Training..." paper. For inference, of course, we somehow need to estimate the segmentation maps.