search

wasidennis / AdaptSegNet

Learning to Adapt Structured Output Space for Semantic Segmentation, CVPR 2018 (spotlight)
845 stars 205 forks source link

Model Structure #60

Closed layumi closed 4 years ago

layumi commented 5 years ago

Dear author,

Thank you for your great code. I find an error in ASPP Classifier. https://github.com/wasidennis/AdaptSegNet/blob/master/model/deeplab_multi.py#L114-L118

    def forward(self, x):
        out = self.conv2d_list[0](x)
        for i in range(len(self.conv2d_list) - 1):
            out += self.conv2d_list[i + 1](x)
        return out  # Remove extra space
wasidennis commented 4 years ago

Yes, this was a bug and was fixed by the reference repo link.

The bug is essentially making our model have only 2 effective ASPP layer, and we don't think it will affect the performance much. Note that, it may require other hyperparameters tuning if fixing this bug in our model.

layumi commented 4 years ago

@wasidennis Thanks a lot.