Mazin-Hnewa / MS-DAYOLO

Multiscale Domain Adaptive YOLO for Cross-Domain Object Detection
MIT License
61 stars 6 forks source link

pytorch yolov5 #5

Closed mrfan98 closed 2 years ago

mrfan98 commented 2 years ago

When I used yolov5 model to train cityScapes' 2975 training sets to test the cityScapes - Foggy validation set, I was always told that no labels were found

Mazin-Hnewa commented 2 years ago

For yolov5, you need to put images and labels in different folders, while here in this repo, we use YOLOv4 which images and labels should be in the same folder.

mrfan98 commented 2 years ago

For yolov5, you need to put images and labels in different folders, while here in this repo, we use YOLOv4 which images and labels should be in the same folder.

What does target set (with dummy annotation) mean when you training, it is have a label or not

Mazin-Hnewa commented 2 years ago

Normally for domain adaptation, labels are not available for target data. The target set does not have any labels, but we put dummy annotations (i.e. empty files) to load target data with yolov4.

mrfan98 commented 2 years ago

Normally for domain adaptation, labels are not available for target data. The target set does not have any labels, but we put dummy annotations (i.e. empty files) to load target data with yolov4.

ok, thanks