HobbitLong / CMC

[arXiv 2019] "Contrastive Multiview Coding", also contains implementations for MoCo and InstDis
BSD 2-Clause "Simplified" License
1.3k stars 179 forks source link

other available views? #31

Open wqtwjt1996 opened 4 years ago

wqtwjt1996 commented 4 years ago

Could you please release the code of using other views instead of only "l" and "ab" in the training CMC process?

HobbitLong commented 4 years ago

Hi, @wqtwjt1996 ,

Which experiment are you referring to, Video, Segmentation and Depth, or other color spaces?

wqtwjt1996 commented 4 years ago

Hi, @wqtwjt1996 ,

Which experiment are you referring to, Video, Segmentation and Depth, or other color spaces?

depth and segmentation.

HobbitLong commented 4 years ago

depth and segmentation.

It needs a bit of work to combine that one in current repo, probably not now.

yassouali commented 4 years ago

@HobbitLong Can you please provide some additional info on how we compute the loss in case of segmentation & depth ? I can't see how to compute the contrastive loss from the segmentation masks, do we pass the labels (C, H, W) through a network to get a representation and then compute the loss? thanks.

HobbitLong commented 4 years ago

@HobbitLong Can you please provide some additional info on how we compute the loss in case of segmentation & depth ? I can't see how to compute the contrastive loss from the segmentation masks, do we pass the labels (C, H, W) through a network to get a representation and then compute the loss? thanks.

As illustrated in the paper, you also need an encoder to encode segmentation masks into features.