pytorch-siamese-triplet icon indicating copy to clipboard operation
pytorch-siamese-triplet copied to clipboard

Published 20 hours ago • verified publisher icon avilash

choosing architectures

Open sezan92 opened this issue 5 years ago • 4 comments

sezan92 avatar Oct 09 '20 10:10 sezan92

Looks good, will test each one and merge. The only doubt I have is the number modules of each architecture. I believe they are same for all the architectures and hence shouldn't be a problem

avilash avatar Oct 11 '20 18:10 avilash

can you explain by "number of modules for each architectures"? i see the training worked after changing the archs to resnet 18 and 34 at least

sezan92 avatar Oct 12 '20 02:10 sezan92

self.features = nn.Sequential(resnet.conv1, resnet.bn1, resnet.relu, resnet.maxpool, resnet.layer1, resnet.layer2, resnet.layer3, resnet.layer4, resnet.avgpool) https://github.com/avilash/pytorch-siamese-triplet/blob/master/model/embedding.py#L12 These are the modules. Should be same across architectures

avilash avatar Oct 12 '20 15:10 avilash

oh yes. they are same

sezan92 avatar Oct 13 '20 02:10 sezan92