Matching between day-to-day and night-to-night

[gguzzy]

Hi dear all, congratulations for your work.

I had one questions, in terms of Day-to-Night and Night-to-Day recognition there is a result table presented in the paper at table 2. My question is, in terms of reassigning the id, what are the results between Day-to-Day (so query in the day and proposed images of same car in day light conditions) and Night-to-Night? Is there any benchmark or mAP related to this? Do you think the model would still reidentify them correcty? If so, why?

Thanks a lot.

[chenjingong]

Dear,

I hope this letter finds you well!

As you see, we didn't focus on day2day or night2night. Actually, we have discussion about this question on time but the result is in this task . Our challenge is focus on cross-domain. But I think if you want try N2N D2D you can try. By the way, there are so many methods already solved D2D. You can get many samples which are mentioned in our paper Introduction part Table1.

I don't know the model still reidentified. But you can see the code. I believe it can be work. Because we have two branches : day-domain and night-domain. So , if you want try D2D and N2N. You can change the branches directly. And in my Semi-Se experiment it can be work, that's why I believe.

Thank you for your interest.

Best Regards, Jingong Chen