Closed ark10806 closed 2 years ago
LB.674
- [LB.603] UniverseNet_r101 - [LB.636] swinB_cascade_rcnn with few pseudo labeled data - [LB.661] swinB_cascade_rcnn with multi-valued-scaling
LB.657
[full]: ready meal, 1.46GB
1.46GB
wget http://pred.ga/AI_Tech/ObjDetection/Datasets/pseudo_dataset_beta1.0_full.zip
[light]: without train & test images, 378.4MB
378.4MB
wget http://pred.ga/AI_Tech/ObjDetection/Datasets/pseudo_dataset_beta1.0_light.zip
swinB_cascade_rcnn으로 확인 결과, pseudo data는 CV 및 LB에서 성능 개선이 없는 것으로 확인되었습니다. 앞으로의 진행 사항은 YOLO v5, k-fold cross validation으로 생각 중입니다.
Teacher models
LB.674
beta 1.0
LB.657
과 10.07(목) 모델LB.674
에서 얻은 pse의 객체 수는 각각 1761, 3161개로 더욱 많은 pse를 확보했습니다.Contents
Download
[full]: ready meal,
1.46GB
[light]: without train & test images,
378.4MB