-
When training a model, we can "ignore" or not worry as much about overfitting as long as the validation error is decreasing.
-
Image Sizes are generally at 224x224 and 299x299, which are the sizes that imagenet models are generally trained at. You get best results if you use the same as the original training size. Since people don’t tend to mention what size was used originally, you can try using both with something like dogs v cats and see which works better. More recent models seem to generally use 299.
-
Rare Cases You can replicate the rare classes to make them more balanced. Never throw away data!
- data augmentation
- pretrained network
- gradually increasing image size
- differential learning rates
- SGDR
- dropouts
- higher resolution images
- When opening a notebook in fastai library, make a copy with the prefix tmp. "tmp" files are included in fastai repo's .gitignore