The number of convolution layer is not enough in models/model_SP.py

[kazucmpt]

You wrote, " The SP module is a set of 5 convolution layer groups following the inverse order of VGG16 while changing all max-pooling layers into upsampling layers." in your paper. In VGG16, the final convolution layer group has three layers. So you have to add one more layer in the first convolution layer group when you decode.

I mean that models/model_SP.py should be modified. You wrote

self.decoder = nn.Sequential( nn.Conv2d(512, 512, kernel_size=3, padding = 1), nn.ReLU(inplace=True), nn.Conv2d(512, 512, kernel_size=3, padding = 1), nn.ReLU(inplace=True), nn.Upsample(scale_factor=2),

But I guess it should be

self.decoder = nn.Sequential( nn.Conv2d(512, 512, kernel_size=3, padding = 1), nn.ReLU(inplace=True), nn.Conv2d(512, 512, kernel_size=3, padding = 1), nn.ReLU(inplace=True), nn.Conv2d(512, 512, kernel_size=3, padding = 1), nn.ReLU(inplace=True), nn.Upsample(scale_factor=2),

Thank you.

[hyf015]

I think either is OK. It wouldn't infect the result too much.

[kazucmpt]

Why do not you modify it?

[hyf015]

I'll modify it soon.