Regarding the COCO Retrieval problem in Table 1

[mop134679852]

Why are the results I get from using the code in FG-CLIP/eval/test.py different from those in your paper? For R@1, R@5, and R@10, the results I get from running your code directly and using your weight .pt file are generally better than the results in Table 1 of your paper.

[lorebianchi98]

Thank you for bringing this to our attention! Could you please share how much of an improvement you're observing in the R@1, R@5, and R@10 metrics compared to the results reported in Table 1 of the paper?

I’ve double-checked by relaunching the evaluation on our side using the same code and weights provided in the repository, and I consistently get results that match the numbers reported in the paper. Let us know so we can investigate further!