LAION-AI / CLIP_benchmark

CLIP-like model evaluation
MIT License
590 stars 75 forks source link

Support computing metrics using test time distribution normalisation #112

Open rom1504 opened 11 months ago

rom1504 commented 11 months ago

https://twitter.com/YifeiZhou02/status/1716513495087472880?t=tktnVDniUhDadYYwEM3F5w&s=19 claims by using a second term in the distance computation they improve image net zero shot classification accuracy by like 2pp on imagenet for B/32 I'm curious if that's true as well for better models