search

Flashkong / Source-Free-Object-Detection-by-Learning-to-Overlook-Domain-Style

Shuaifeng Li, Mao Ye, Xiatian Zhu, Lihua Zhou, Lin Xiong, Source-Free Object Detection by Learning to Overlook Domain Style, CVPR2022-Oral paper, 2022.
MIT License
49 stars 6 forks source link

Nice work #9

Open ross-Hr opened 1 year ago

ross-Hr commented 1 year ago

Hi, Can upgrade python to 3.7+ ?

Flashkong commented 1 year ago

Thank you so much for your appreciation of our work!

Our work is based on jwyang/faster-rcnn.pytorch, which requires python version 2.7 or 3.6 and pytorch version 1.0. I think python3.7+ may be feasible, but I suggest you use pytorch1.0.0 to meet the requirements of jwyang/faster-rcnn.pytorch. Note that we haven't tried it, so there's no guarantee it will work.

In addition, please ensure that the source domain model and the target domain model use the same environment for training.

simvarail commented 1 year ago

Thank you so much for your appreciation of our work!

Our work is based on jwyang/faster-rcnn.pytorch, which requires python version 2.7 or 3.6 and pytorch version 1.0. I think python3.7+ may be feasible, but I suggest you use pytorch1.0.0 to meet the requirements of jwyang/faster-rcnn.pytorch. Note that we haven't tried it, so there's no guarantee it will work.

In addition, please ensure that the source domain model and the target domain model use the same environment for training.

Hi, Have you attempted to apply your method to other detectors like YOLO? Do you believe it could be used with YOLO, considering that YOLO lacks an Region Proposal Network?

Flashkong commented 1 year ago

Our work provides a novel idea for SFOD by overlooking target domain style for model adaptation. Theoretically, this "overlooking domain style" idea can be applied to any type of detector, whether it is single-stage or two-stage, or even transformer-based detectors. However, it is important to note that the application based this idea may vary depending on the specific detector and should be tailored accordingly.