search

TorchDSP / torchsig

TorchSig is an open-source signal processing machine learning toolkit based on the PyTorch data handling pipeline.
MIT License
155 stars 37 forks source link

No Documentation for Creation of All Pretrained Models Found at https://torchsig.com/dist/downloads.html #206

Closed sciafri closed 2 days ago

sciafri commented 1 year ago

Describe the bug After creating the datasets, I would like to recreate the pretrained models. However, it's not clear how to do so for all datasets and model combinations listed at: https://torchsig.com/dist/downloads.html. Could you please describe where I might be able to find this information to generate all pretrained models listed on https://torchsig.com/dist/downloads.html ?

After a quick read through of the code, I don't see where any of that is documented. There are two places which hint at somethings missing in the repository:

  1. scripts/train_sig53.py only details how to train efficientnet on sig53. Do I use this script as a template and modify this script for each of the pretrained models?

  2. As it is right now, models/iq_models and models/spectrogram_models only lists efficientnet, xcit, and detr. Yet on the downloads there are additionally: YOLOv5, Mask2Former, PSPNet. How are those pretrained models supported by the code? Did you simply just modify the train_sig53.py file to have those models in the template? I want to make sure I'm getting exactly the pretrained models created on that website, so confirmation about this would be helpful.

Thank you.

MattCarrickPL commented 2 days ago

We are in the process of updating the set of pretrained models with new ones and much of the documentation about how to train models is within the new example notebooks