Parameters of BN layers in branch models seem not be frozen during inference

siqueira-hc / Efficient-Facial-Feature-Learning-with-Wide-Ensemble-based-Convolutional-Neural-Networks

Efficient Facial Feature Learning with Wide Ensemble-based Convolutional Neural Networks

MIT License

193 stars 41 forks source link

Parameters of BN layers in branch models seem not be frozen during inference #22

Open siyilingting opened 3 years ago

siyilingting commented 3 years ago

Dear author，thanks for your sharing. A problem on Line210 in esr_9.py confused me. When loading the branch models, the parameters of the BN layers did not seem to be frozen, which caused output of self.bn1(self.conv1(x_shared_representations)) (Line 108 in esr_9,py) and the prediction results to be changed when the photos were tested in batch. Would this problem affect the performance of the model? Look forward to your favourable reply.