A question about the benchmark results. about dropmae HOT 4 CLOSED

Plus, it's not a big deal since it's not SOTA anymore, but I still hope to know so that I can fix the underlying bugs in the open-source work accordingly. Please kindly shoot me a message if you have time.

from dropmae.

FYI, the original results can be found here.

from dropmae.

Hi Shenyuan,
Thanks for your comments. I just checked the AIATrack results on LaSOT in Table 2, yes, it should be a typo. For LaSOT_ext, I use Table 3 in your paper w/ the variant of "AiA in both" (46.8/54.4/54.2) for comparison , which is slightly different from the results reported in your git. What's the diff?
Anyway, I will fix the typos in the arxiv version.

from dropmae.

I see, thanks for your quick reply and generous help. I totally understand for such a solid work this is just a minor issue. From Table 5 in AiATrack paper, you will find that I have done some parameter tuning to push the performance. Thus, the publicly released raw results on GitHub are under the best settings, i.e., LaSOT (69.0/79.4/73.8) & LaSOT_ext (47.7/55.6/55.4). You are free to verify my results via running the evaluation. I sincerely appreciate your help if you can update the results accordingly. BTW, you guys are so rich with an enviable amount of GPU resources, hahaha :)

from dropmae.