Redistribution notice: pretrained AdaCLIP weights mirrored on Hugging Face (cubert-gmbh/adaclip)

[nghorbani]

Hello, and thank you for releasing AdaCLIP.

We (Cubert GmbH, https://github.com/cubert-hyperspectral) use AdaCLIP in our open-source
hyperspectral toolkit Cuvis.AI (plugin: https://github.com/cubert-hyperspectral/cuvis-ai-adaclip).
Our users kept failing to fetch the pretrained weights from the Google Drive links in the README
(quota and HTML-interstitial errors), so we have mirrored the three released checkpoints, unchanged,
on Hugging Face:

https://huggingface.co/cubert-gmbh/adaclip

• pretrained_all.pth, pretrained_mvtec_colondb.pth, pretrained_visa_clinicdb.pth, exactly as
  downloaded from the Drive ids in the README (sha256 values are listed in the model card)
• your MIT LICENSE file, verbatim
• a model card that credits the AdaCLIP paper and repository, states that the weights were released
  by you without a separate licence, notes the training data (MVTec AD, VisA, ClinicDB, ColonDB) and
  links back here

One thing we noticed while doing this: the README table labels the Drive id 1xVX... as
"MVTec AD & ClinicDB" and 1QGm... as "VisA & ColonDB", but the files behind them are named
pretrained_mvtec_colondb.pth and pretrained_visa_clinicdb.pth. We kept the filenames as they
are and mention the mismatch in the card; if you can tell us which reading is correct we will update
the card.

If you would prefer that the weights are not mirrored, or want a different licence statement or
attribution, please say so here and we will take the repository down or change it right away.