Feature/enhance training #4
Conversation
Irynei commented Oct 4, 2018
Irynei commented - Add DenseNet121 config.
- Add test process.
- Add early stopping.
- Refactor.
| return out | ||
| | ||
| | ||
| class DenseNet(nn.Module): |
There was a problem hiding this comment.
i think we could use a model from torchvision
There was a problem hiding this comment.
Not for Cifar10. Models from torchvision are pre-trained on ImageNet and hence design to work with 224x224 input. In fact, it is stated in their documentation that images are expected to be at least of size 224x224.
I tried densenet121 from torchvision with Cifar10 and it didn't work.
There was a problem hiding this comment.
you can use a constructor (e.g. torchvision.models.densenet.DenseNet), not a particular implementation.
There was a problem hiding this comment.
There is still some issue with AvgPool2d. (e.g. pytorch/vision#155, https://discuss.pytorch.org/t/torchvision-resnet-input-size/9405/2)
I will investigate how it could be done efficiently.
