The usage of masks

[YibinXie]

Hi Authors,

Thanks for this great repo. I just wonder what is the purpose of using masks in COCO dataset?

I have tracked the code for masks. My guess is that masks are used for shrinking the potential area for possible keypoint positions when calculating loss, because keypoints are meant to be inside one's segmentation mask. COCO is a richly labeled dataset, I mean it has keypoint labels along with segmentation masks, so we can use the mask for assisting in calculation loss. And we surely can cancer masks out, which may have a relatively large number of loss.

Am i right? Please correct me if i were wrong. Looking forword to your reply.

Regards and thanks, Yibin

[cucdengjunli]

same question