Is there no other loss in the part of "Image Feature Extractor" ?

[wl03036530]

Hi,
I want to know how to get 2D pose, segmentation and dense mapping encoding without any loss in "Image Feature Extractor" ?
I also saw that there are three folders: dense, hms, and mask in the processed dataset folder. If it is not used here, which side needs these data?

thanks.

[wl03036530]

And another question.
Is it convenient to release loss code or the annotation after processing from InterHand2.6M?
Because I found that the order of the index of joint seems to be different from the original. (ex: on the line 128, 129 of eval_interhand.py. Root index is 9 but for original index is 20, or somewhere I missed?)

thanks.