Transfer Learning #1178
-
|
First of all, thanks for this repository. In general terms, it is very useful. Is it possible to use this repo for a custom transfer learning approach? Let's say that I trained a model using this repo and I want to use this trained model as a feature extractor. Then, I will retrained a new fully connected layer attached to the previous feature extractor. Is there something implemented in this sense? Any guidance regarding to this discussion will be fully appreciated. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
@bryanpiguave timm models can be used like any torchvision model for fine-tuning (transfer) from pretrained models.... there are additional features in the interface to make it even easier to deal with changing the classifier layer consistently across all models... some info in various guides https://github.com/rwightman/pytorch-image-models#getting-started-documentation |
Beta Was this translation helpful? Give feedback.
@bryanpiguave timm models can be used like any torchvision model for fine-tuning (transfer) from pretrained models.... there are additional features in the interface to make it even easier to deal with changing the classifier layer consistently across all models... some info in various guides https://github.com/rwightman/pytorch-image-models#getting-started-documentation