VDIGPKU / T-SEA

[CVPR 2023] T-SEA: Transfer-based Self-Ensemble Attack on Object Detection
93 stars 10 forks source link

About pretrained patch #26

Closed Bomingmiao closed 2 months ago

Bomingmiao commented 2 months ago

Hi, your code is really wonderful. Could you please offer your pretrained AdvPatch on these models ? I would really appreciate it if you kept that and provide me with them.

ziyannchen commented 2 months ago

The pretrained AvdPatch has already offered in GoogleDrive / BaiduCloud, please refer to the patch zoo in the README file

Bomingmiao commented 2 months ago

The pretrained AvdPatch has already offered in GoogleDrive / BaiduCloud, please refer to the patch zoo in the README file

Yes, I have downloaded the files from both GoogleDrive and BaiduCloud. You have offered all of the patches of T-SEA. But as for AdvPatch, which is the baseline and the competitive method in your paper, there is only one patch trained on YOLO v2. If you kept patches trained on the other models with the method of AdvPatch, can you share them with me?

ziyannchen commented 2 months ago

Hey, AdvPatch is provided by the original paper which you can find in their official ropo. Note that we use the patch named object_score.png whose training objective is consistent with ours.