search

kienduynguyen / BoxeR

Code release for "BoxeR: Box-Attention for 2D and 3D Transformers"
MIT License
140 stars 22 forks source link

Evaluation Results AP=0 #24

Open 351246241 opened 1 year ago

351246241 commented 1 year ago

Thank you for share this work! I use python tools/run.py --config e2edet/config/COCO-InstanceSegmentation/boxer2d_R_50_3x.yaml --model boxer2d --task detection fortraining in coco dataset. then,I evaluate coco dataset with python tools/run.py --config e2edet/config/COCO-InstanceSegmentation/boxer2d_R_50_3x.yaml --model boxer2d --task detection training.run_type=val. But I get AP =0,like this image I assumed it was a training error on my part and then downloaded the provided pre-training weights. But got the same result python run.py --config e2edet/config/COCO-InstanceSegmentation/boxer2d_R_101_5x_lsj_resmode_v2.yaml --model boxer2d --task detection training.run_type=val image

kienduynguyen commented 1 year ago

Sorry for the confusion! You should also specify the model path for the pre-trained checkpoint:

python run.py --config e2edet/config/COCO-InstanceSegmentation/boxer2d_R_101_5x_lsj_resmode_v2.yaml --model boxer2d --task detection training.run_type=val training.resume=True training.resume_file=