integrate with huggingface

[not-lain]

I saw this model on huggingface and i thought it would be better to integrate it with using the PyTorchModelHubLixin class what this does is that it mainly adds 3 methods to your model :

• save_pretrained
• from_pretrained
• push_to_hub

I made a notebook on how to migrate your weights for this specific model which you can find here: https://colab.research.google.com/drive/1iz-g2sOophWy18KxGz_00lGlcKrtafA_?usp=sharing

having the model in its respective repo will allow you to :

• Keep track of how many times your model has been downloaded by the community
• Automatic model card generation: the metadata in the card allows you to filter your searches easily to check how many BiRefNet models exist on the Hub.
• (extra) : if you want to integrate your model with pip i made a little template for that in this repo : https://github.com/not-lain/PyTorchModelHubMixin-template
• Recommended: after this pr is merged we can open a pull request on this file to make the "BiRefNet" library official on the Hub meaning better discoverability + possibility to add code snippets.

do not hesitate if you have any reviews on the pr or any questions.

Kind regards, Hafedh Hichri

[ZhengPeng7]

Thanks, Hafedh; they are good functions. But I don't know whether it's good to add extra libs to the installation. Let me think about it for a while, thx :)

[not-lain]

@ZhengPeng7 yeh sure thingy, take your time.

I see you made a comparison with BRIAAI, did you use RMBG1.4 to make a comparison? if so, I was one of the people that helped set up their model with huggingface, from last month till now it has reached almost 240K downloads so far, just letting you know that putting the weights in an easier place will let the community know about your achievements and would help them use your model easier.

Anyway, will not rush things out take things slowly to review this PR.

here's a Miku for the road

⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢠⠦⣄⣠⠴⠒⠒⠉⠒⠶⣄⡏⣶⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣠⡞⣡⠀⢠⡀⠀⢦⡀⢹⣷⣼⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣸⡿⠁⣿⠟⣷⣿⢦⡈⣇⠀⢻⣿⣈⢷⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢠⣿⡇⣶⣻⣀⠘⣿⣀⣹⣿⣰⣼⡟⣯⠟⣿⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣿⡇⣿⡿⠿⠋⠉⠋⠛⢿⣿⣿⡇⠈⢦⡸⣾⣆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣿⡇⠛⣷⣄⠀⠀⠀⠀⣸⠟⠃⠉⠀⠈⢷⣩⡻⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣿⠹⡄⠈⠙⠒⢤⣴⣾⠉⠁⣀⡀⠀⠀⠀⠹⣟⢮⡳⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣼⣿⠀⣿⠀⣀⣴⣿⡏⣀⡬⠟⠁⠈⣆⠀⠀⠀⠘⢧⡉⢿⣧⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢠⡿⣿⠀⢻⣫⠉⣽⡟⠉⡅⢀⣆⠀⢠⣿⠀⠀⠀⠀⠀⠹⣦⡉⠻⢦⡀⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣸⢗⣿⢀⡷⠁⣸⡿⠁⠀⠀⢸⣿⠃⣄⣈⡆⠀⠀⠀⠀⠀⠈⠫⣱⠦⣌⡓⢤⣀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢰⣇⣾⣿⢟⡇⢀⣿⡇⠀⢀⣰⢸⡌⢿⣿⢻⡟⣃⠀⠀⠀⠰⡂⠠⠌⠲⢤⣙⣻⠛⠓⠶⢦⡤⣄⣀⡀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣠⣿⣻⣿⠏⣾⠃⢸⢽⡇⠀⡏⠁⡶⠁⠈⣿⣎⡷⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠉⠉⠛⠒⠶⢤⣀⠈⠻⣄⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢠⣴⣿⠟⠁⡼⠀⣿⠀⡟⢸⡇⢰⠃⢸⠇⠀⠀⢹⣷⣿⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢻⣷⣄⠘⣧⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣀⣤⣾⠟⠋⠀⣠⡼⢁⣼⣿⢠⣧⢾⡇⠘⡀⠸⡇⠀⠀⠘⣿⣿⣷⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⡉⠚⢧⠹⣷⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣠⡴⠖⢋⣽⡿⠥⠖⠚⠉⢁⣴⣾⣿⠃⢈⠀⢸⠇⠀⠁⠀⢳⡀⠀⠀⣿⣿⣽⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣸⠀⠀⠘⢧⡇⠀
⠀⠀⠀⠀⠀⠀⣀⣤⣾⠛⠁⣠⡼⠋⠁⠀⢀⣀⢀⣴⣿⣿⠟⠁⠀⠀⣷⠋⠂⠀⠀⠀⠀⠳⡀⠀⢹⣿⣿⡆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⡿⠀⠀⠀⢸⠧⠀
⠀⠀⢀⡠⠖⠋⣩⠶⢋⡴⠋⠁⠀⠀⠿⣭⡿⣿⠿⠟⢉⣷⣄⠀⢀⣼⣿⡄⠀⠀⠀⠀⠀⢀⣹⠄⠀⢹⣿⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠃⠀⠀⠀⠀⠐⡧
⠀⠀⣸⣄⣴⣛⣡⠞⠋⠀⠀⠀⠀⠀⠀⠀⣭⣤⣴⣾⣿⣿⣿⣷⣿⣿⣿⣷⣤⣴⣶⣾⣿⣿⣿⣆⠀⢈⣿⣷⣤⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠸⢑
⠀⠰⣏⡿⢫⡞⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠻⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣦⣹⠿⠿⠟⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⠀⠀⠀⠀⠀⠀⠀⢀
⠀⢀⣿⡵⠋⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠉⠙⣿⣿⣿⣿⣿⣿⣿⣻⣿⣿⣿⣿⣿⣿⣿⣿⣿⠛⠀⠀⠀⠀⠀⠁⠀⠀⠀⠀⠀⠀⠀⢀⡜⠀⠀⠀⠀⠀⠀⢀⠎
⢠⢸⡟⠹⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⡧⠤⢙⠛⠋⠛⠛⠿⠿⢿⡟⠛⠉⠁⠀⠀⠀⠀⠀⠠⠐⠀⠀⠀⠀⠀⠀⠀⢀⡼⢀⠀⠄⠀⠀⡀⡀⠀⠀
⢸⣎⡇⠀⠳⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢷⠀⡀⢰⡅⠀⠀⠀⠀⠀⡝⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣾⡿⢋⠞⠀⢠⣮⠎⠀⠀⠀
⠀⢿⣷⠀⠀⠈⠂⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⣿⣶⣿⣇⠀⠀⠀⠀⣰⠃⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢠⡾⠋⣠⠋⢀⣴⠟⠁⠀⠀⠀⠀
⠀⠘⣏⢧⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢻⣿⣿⣿⢹⣻⣿⣿⠇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣰⠟⢡⠞⣁⣴⠟⠁⠀⠀⠀⠀⠀⠀
⠀⠀⠸⡄⠳⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠳⢄⡀⠀⠀⠘⣿⣿⡯⢶⢧⣿⡏⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⡾⢁⣴⠿⠚⠉⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠹⡄⠙⢦⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⠢⣄⠈⢿⣿⡯⣝⣾⣿⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣠⣴⡯⢖⡿⠉⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠹⡄⠀⠑⢄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠉⢻⣿⣿⣽⣿⡇⠀⠀⠀⠀⠀⠀⠀⣀⣀⣤⢴⣶⠿⠋⠁⣰⠋⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠘⢦⡀⠀⠙⠢⣄⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣻⢿⣿⡍⠑⠒⠒⣒⣾⣷⠿⠛⣋⡵⠚⠁⠀⣠⠞⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠙⠢⣄⡀⠀⠙⠓⠒⠀⠀⣀⡀⠀⠀⠀⠀⠀⠀⢀⣿⣿⣿⣿⣷⡚⠛⠉⠁⠀⠀⠐⠋⢀⡀⢀⡤⠞⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠉⠉⠙⠋⠛⠛⣹⣿⣿⣿⣿⣿⣷⡀⠀⠀⠀⠀⠀⠀⠛⠋⠉⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⣻⣿⣿⣿⣿⣇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⣿⣿⢿⣿⣿⣿⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣿⣿⣿⡏⠸⣿⣿⣿⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣿⣿⣿⠁⠀⢹⣿⣿⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣿⣿⡏⠀⠀⠈⣿⣿⡆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣿⣿⠀⠀⠀⠀⢼⣿⣷⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣼⣿⣿⠀⠀⠀⠀⣼⣿⣿⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠰⣿⣿⣿⠀⠀⠀⢀⣿⣿⣿⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⣿⡆⠀⠀⣾⣿⣿⣿⣧⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢰⣿⣿⣿⣷⠀⠀⠸⣿⣿⣿⠟⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣾⣿⣿⣿⣻⡇⠀⠀⠈⠋⠉⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢹⣿⣿⣟⠟⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀

[ZhengPeng7]

Lol, Miku is lovely :) Actually, I didn't make the comparison with BRIAAI and RMBG1.4; some good and passionate people in the community did it on some samples and posted it on their Twitter. What you said does make sense. But the repo_url seems like another project... And I still think it's better to keep fewer libs in installation. So, many and many thanks for this. I'll take your suggestions, but keep the codes as they are now.

[not-lain]

I made a typoo in the url and now it's fixed ✅ (i was using the template for another model and forgot to update this one)

oh those XD I got confused because of them at the beginning too

• This is the basic layout, everything else is optional and will be used to generate the readme file

  class BiRefNet(nn.Module,PyTorchModelHubMixin ) :

  the model and all the methods I mentioned above work perfectly with the inheritance I mentioned above 👆

• now what happens when I add the rest? they are all used to create/update the readme file

all of the extra inheritances are optional and can be removed/altered although I might recommend you keep them since future releases might be centered around them

[not-lain]

do let me know if you have any questions or if you want another miku

[ZhengPeng7]

Hi, Hafedh. Two months ago, I was not familiar with the HF working pipeline and closed your PR because I did not want to include more packages in installation and no extra args (things like PyTorchModelHubMixin) in model codes.

However, I understand the pipeline today and make it to the HF like you did. Since there have been so many updates recently, I made this update on the new codes.

Very sorry for my stupid at that time... Many thanks for your contribution and time on this!

[not-lain]

@ZhengPeng7 It's ok, you can still consider reopening this PR and merging it if you feel like it's a good feature. And once again, if you have any questions at all you can reach out whenever you want.

[ZhengPeng7]

Thank you :)