search

naver-ai / pcmepp

Official Pytorch implementation of "Improved Probabilistic Image-Text Representations" (ICLR 2024)
https://naver-ai.github.io/pcmepp/
Other
51 stars 1 forks source link

Can you provide the detailed results on noisy COCO? #2

Closed leolee99 closed 1 year ago

leolee99 commented 1 year ago

Hi! Thanks for your excellent work. I'd like to compare with some baselines trained on noisy COCO set and cite the results in Table 2. But only average R@1 performances of COCO 1K and 5K are reported. Can you provide more detailed R@K performance on both i2t and t2i?

Sincerely thanks for your help!

SanghyukChun commented 1 year ago

Hi, thanks for your interest! I am now working on rechecking all the experiments in the paper, and some numbers will be updated. It would take about 2-3 days. If you need the numbers immediately, let me check Table 2 first.

leolee99 commented 1 year ago

Thanks for your help! I only need the result numbers.

SanghyukChun commented 1 year ago

@leolee99 Hello, thanks for your patience I made a spreadsheet to show the full numbers. (it is still under update) (See Noisy Correspondence (avg) for your request) Feel free to ask me anything if you have any follow-up questions https://docs.google.com/spreadsheets/d/1EMbp03uD7X5puN80zV2wr43Z2FWC4wQ9bo8HXcyU27k/edit?usp=sharing

leolee99 commented 1 year ago

Thanks for the kind author and your excellent work!