about the size of prototype. about proda

[tudragon154203]

Hi ShifuShen,

I don't belong to the author team, but I might know the answer for your confusion.

The feature is extracted right before the last layer of Classifier_Module2 in deeplabv2.py. If you look at the forward function of Classifier_Module2, you might see that:
out['feat'] is input of layer nn.Conv2d(256, num_classes, kernel_size=1, padding=0, dilation=1, bias=False)

So it has 256 kernels in the end.

[ShifuShen]

Hi ShifuShen,

I don't belong to the author team, but I might know the answer for your confusion.

The feature is extracted right before the last layer of Classifier_Module2 in deeplabv2.py. If you look at the forward function of Classifier_Module2, you might see that:
out['feat'] is input of layer nn.Conv2d(256, num_classes, kernel_size=1, padding=0, dilation=1, bias=False)

So it has 256 kernels in the end.

Got it Thanks ~