victor000000

for the stride 2 conv in class SPP_NET, we have to ensure that the size of tensor > 26 * 26 otherwise, the kernal(4 * 4) of self.conv5 will larger...

while in the test time, what dose the _warmup_batchnorm do?

1

about the function _warmup_batchnorm in utils.py I guess its used for re-evaluate the parameters of the last layer in Encoder(in model.py) as the default "Self-Supervised Training", use_bn is false 1....