Dataset and training of pre-trained models in pytorch vision

vision
1

I’m using te pre-trained models available here: Models and pre-trained weights — Torchvision main documentation

I would like to be sure about the dataset used for training those models (for example, imagenet with 1k categories) and he training procedure (for example, if images are normalized to a standard size before training, etc).

Where can I find this information?

2

Two relevant sources on this matter:

  1. https://pytorch.org/vision/stable/training_references.html, training scripts references
  2. resnet50 — Torchvision main documentation, where you can see preprocessing steps. The same type of info is available for other models

Using the latter, I was able to reproduce some well-known finetuning scores based on resnet50 efficiennet L, etc