search

forresti / SqueezeNet

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters
BSD 2-Clause "Simplified" License
2.17k stars 723 forks source link

SqueezeNet training on cifar #54

Closed GitZinc closed 6 years ago

GitZinc commented 6 years ago

The accuracy is 0, and the loss is too high all the time when I run the model on cifar10. Do I need to delete avg pooling layer?

morawi commented 6 years ago

I am having the same issue on cifar10. The accuracy is chance level (10%), no progress is achieved at all.

morawi commented 6 years ago

Well, now I am getting the SqueezeNet to move (I am using PyTorch): The problem was using an ImageNet pre-trained model, so setting pre-trained-model to False solved the problem.

GitZinc commented 6 years ago

Wow, I used caffe at that time. Thanks for your reply, now I think I find the problem.