maybe it is an error

[1119066022]

hello,when i am learning your code,i find that your code named fpn.py may has an error:

Semantic

    _, _, h, w = p2.size()
    # 256->256
    s5 = self._upsample(F.relu(self.gn2(self.conv2(p5))), h, w)

the size there is directly upsampled from 1/32 to 1/4,so maybe your upsample function should use :
def _upsample(self, x):
return F.interpolate(x, scale_factor=2, mode='bilinear', align_corners=True)
then:
_, _, h, w = p2.size()
# 256->256
s5 = self._upsample(F.relu(self.gn2(self.conv2(p5))))