search

ultralytics / xview-yolov3

xView 2018 Object Detection Challenge: YOLOv3 Training and Inference.
https://docs.ultralytics.com
GNU Affero General Public License v3.0
252 stars 57 forks source link

class weight hard mining #19

Closed yangxu351 closed 4 years ago

yangxu351 commented 4 years ago

Hello, I know this repo is not under active, but I want to reproduce the repo on xview, could you please explain how did you get the hard mining class weight in utils/utils.py? Screenshot from 2019-12-25 17-18-28

glenn-jocher commented 4 years ago

@yangxu351 inverse frequency weights.

yangxu351 commented 4 years ago

@glenn-jocher Thanks for your quick reply. But I feel a little confusing, the xview class weights are obtained according to the inverse frequency of each class, are the xview_class_weights_hard_mining also obtained according to the inverse frequency weights? What's the difference? Or anything I wrongly understood? 2019-12-27 (5)

glenn-jocher commented 4 years ago

@yangxu351 this may refer to image or chip weighting, which deals with how often to present a given image or chip for learning. What you are showing above are the class weights, which are inverse frequencies.

glenn-jocher commented 4 years ago

@yangxu351 in any case, I do not recommend using this repo, as https://github.com/ultralytics/yolov3 is much better performing and currently maintained. You can train COCO from scratch to 61.4 mAP for example on it, better than darknet.

yangxu351 commented 4 years ago

@glenn-jocher Thanks.

github-actions[bot] commented 4 years ago

This issue is stale because it has been open 30 days with no activity. Remove Stale label or comment or this will be closed in 5 days.

glenn-jocher commented 10 months ago

@yangxu351 you're welcome! If you have further questions, feel free to ask.