Hello, thank you very much for sharing this great work!
I'm in the process of testing other Dino based models on this benchmark. I noticed that for the SPair dataset we're doing a resizing of images while also in the evals/models/dino.py we're doing a centred padding.
I was testing the benchmark on different resolutions and noticed a difference in performance when both resizing + padding is being enforced compared to just resizing.
I was wondering what is the purpose of using both (for base SPair experiments I believe we're resizing images to 800 x 800 and then applying centred padding) or in general whether resizing/centred padding was preferred during evaluation. Thank you!
Hello, thank you very much for sharing this great work!
I'm in the process of testing other Dino based models on this benchmark. I noticed that for the SPair dataset we're doing a resizing of images while also in the evals/models/dino.py we're doing a centred padding.
I was testing the benchmark on different resolutions and noticed a difference in performance when both resizing + padding is being enforced compared to just resizing.
I was wondering what is the purpose of using both (for base SPair experiments I believe we're resizing images to 800 x 800 and then applying centred padding) or in general whether resizing/centred padding was preferred during evaluation. Thank you!