MobileNetV2-pytorch

MobileNetV2-pytorch copied to clipboard

[number of classes](https://github.com/Randl/MobileNetV2-pytorch/blob/master/imagenet.py#L117), here the net use 1000 as default, this line code, I suggest that, we can use `len(train_loader.dataset.classes)`.

zacario-li

When training like this : python imagenet.py --dataroot "/home/liming/data/" --resume "results/mobilenet_v2_1.0_224/model_best.pth.tar" -e I got: Missing key(s) in state_dict: "conv1.weight", "bn1.weight", "bn1.bias", "bn1.running_mean".......

Daonancai

Duplicate log_softmax

1

Looking throughh the source code for torch.nn.CrossEntropyLoss(), it seems to be doing log_softmax inside it. The model is also doing a log_softmax at the end. Isn't this a duplicate? Is...

mathmanu

How did you compute a metric like images/second?

1

I am trying to use standardized metric such as images/sec to arrive at relative training speeds for different frameworks(e.g. PyTorch and TF). Have you computed such a metric, if not,...

skalyan

Question: loss value is a little large.

3

Hi, dear: From here: https://github.com/Randl/MobileNetV2-pytorch/blob/master/results/mobilenet_v2_0.5_160/results.csv I can see the loss value is still a little large for a good module, from loss definition, loss should be between 0-1, and when...

asimay