search

xrli-U / MuSc

This is an official PyTorch implementation for "MuSc : Zero-Shot Industrial Anomaly Classification and Segmentation with Mutual Scoring of the Unlabeled Images" (MuSc ICLR2024).
MIT License
207 stars 12 forks source link

Will the relevant training code be made public in the future? #2

Open wuyefa opened 3 months ago

wuyefa commented 3 months ago

Hello! This is really a good piece of work. Will the related training and fine-tuning code be open sourced in the future? The data of our usage scenario is relatively special, and we want to try the effect of fine-tuning. Thank you so much.

wuyefa commented 3 months ago

Sorry! It's an amazing work！ I thought the visual parameters of the clip used were fine-tuned.

xrli-U commented 3 months ago

Hello! This is really a good piece of work. Will the related training and fine-tuning code be open sourced in the future? The data of our usage scenario is relatively special, and we want to try the effect of fine-tuning. Thank you so much.

As a zero-shot method, our MuSc uses a pre-trained image encoder of CLIP as the feature extractor, requiring no additional training and fine-tuning. Our code will automatically download the pre-trained model, or you can download the pre-trained ViT of CLIP and DINO via the official projects below. CLIP: https://github.com/mlfoundations/open_clip DINO: https://github.com/facebookresearch/dino DINO_v2: https://github.com/facebookresearch/dinov2

wuyefa commented 3 months ago

This is really amazing. I initially thought it would require at least fine-tuning on industrial datasets. Thank you for your explanation.