Why the evaluation will change while you evaluate the model at different time?Anyone knows what happened? PLEASE HELP! about abd-net HOT 2 CLOSED

You missed some parameters. See #21

from abd-net.

You missed some parameters. See #21

I also using flip-eval while evaluation. Did you notice the network changing sometime, these two parts below often exchange :
(1): GlobalBranch(
(fc): Sequential(
(0): Linear(in_features=960, out_features=1024, bias=True)
(1): BatchNorm1d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace)
(3): Dropout(p=0.5)
)
(avgpool): AdaptiveAvgPool2d(output_size=1)
(classifier): Linear(in_features=1024, out_features=333, bias=True)
)
and
(sum_conv): Sequential(
(0): Dropout2d(p=0.1)
(1): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1))
)
(avgpool): AdaptiveAvgPool2d(output_size=1)
(classifiers): ModuleList(
(0): Linear(in_features=1024, out_features=333, bias=True)
(1): Linear(in_features=1024, out_features=333, bias=True)
)
)

from abd-net.