search

deeplearning-wisc / MCM

PyTorch implementation of MCM (Delving into out-of-distribution detection with vision-language representations), NeurIPS 2022
72 stars 9 forks source link

Model weights for fine-tuning #10

Open jimo17 opened 1 year ago

jimo17 commented 1 year ago

Hi, thank you very much for open sourcing. The comparison of MSP, Energy, and MOS methods is involved in Table 2. of your paper. I want to learn from your thesis method. Is the weight file of the CLIP model that has been fine-tuned for these methods still retained? If these files are still saved, can you share them with me? Thank you very much

black0017 commented 10 months ago

I am not an author of this paper nor affiliated.

It seems that MSP and energy are computed based on a CLIP image encoder after linear probing. I don't think they fine-tuned CLIP in this work. Fort et al. require full training/fine-tuning + Mahalanobis distance as an OOD score.