modelv1 1, 16, 256
modelv2 0, 16, 256
modelv3 stack noise with only 1*1 conv layer and add 2 more 3*3 conv
modelv4 use rep - delta
modelv5 stack noise with only 1*1 conv layer
- modelv5l1 stack noise with only 1*1 conv layer with loss weight 1 (loss = imgage loss + loss weight * sparsity loss)
- modelv5l01 stack noise with only 1*1 conv layer with loss weight 0.1
- modelv5l001 stack noise with only 1*1 conv layer with loss weight 0.01
modelv6 stack noise for all layer