search

justinkay / aldi

Official implementation of "Align and Distill: Unifying and Improving Domain Adaptive Object Detection"
https://aldi-daod.github.io/
38 stars 7 forks source link

Typo in the table 3? #28

Closed sshhj89 closed 1 month ago

sshhj89 commented 1 month ago

Hi, I am really interested in your paper, research, and company. I would like to ask about the paper. In page 13 and table 3 it shows the ablation study using L_align and L_distillation. It seems like not matching with "we show that feature alignment used in isolation (i.e. without self-training) offers performance gains up to 2.6 AP50. However, these performance gains are smaller than those seen from self-training (AP50FCS of 61.7 vs. 63.1, respectively)."

In the table 3 when you applied L_distillation (a.k.a. self-training), it is 63.7 AP50. Am I wrong?

justinkay commented 1 month ago

You're right! Thank you for pointing this out. It should say 63.7. We will fix this.