search

mysee1989 / TCAE

Self-supervised Representation Learning from Videos for Facial Action Unit Detection
162 stars 27 forks source link

Having trouble understanding maxpercentile, minpercentile, w1, w2... #17

Open JianSShi opened 3 years ago

JianSShi commented 3 years ago

Hi, I am trying to read the code, in the self_supervised_train_TCAE, there has "maxpercentile" and "minpercentile", which seems to narrow the region of the image, but I not sure why we need that and what are they aiming for?

Also, can anyone help me understand how the weight "w1-w10" value has been set as w1 = 10 w2 = 1 w4 = 0.1...

zzx820302704 commented 3 years ago

@JianSShi Have you downloaded the voxcele1/2 datasets? Can you share it? It's too big

JianSShi commented 3 years ago

@JianSShi Have you downloaded the voxcele1/2 datasets? Can you share it? It's too big I did not use that dataset, sry

anxiaoci commented 3 years ago

@JianSShi Have you downloaded the voxcele1/2 datasets? Can you share it? It's too big I did not use that dataset, sry Which dataset do you use?Could you please share the method? I have been trying some dataset else,but could not run this code successfully

JianSShi commented 3 years ago

@JianSShi Have you downloaded the voxcele1/2 datasets? Can you share it? It's too big I did not use that dataset, sry Which dataset do you use?Could you please share the method? I have been trying some dataset else,but could not run this code successfully

I randomly choice some videos to create my own dataset. Basically, all you need to do is get few set of images that each set images are source image and target image. Then, in the TCAE_data.py file, make sure you modify the getitem as how your dataset are formed.