Skip to content

Latest commit

 

History

History
105 lines (70 loc) · 2.99 KB

EVALUATION.md

File metadata and controls

105 lines (70 loc) · 2.99 KB
Raw

Evaluation

We take the checkpoint video-joint_r50.pth as an example. We have listed all the evaluation datasets in the config files.

For the swin-L backbone, please change the --config-file and MODEL.WEIGHTS arguments.

RIS

python3 projects/UniRef/train_net.py --config-file projects/UniRef/configs/eval/r50/eval_rec_r50.yaml --num-gpus 8 --eval-only MODEL.WEIGHTS video-joint_r50.pth

FSS

1-shot evaluation.

python3 projects/UniRef/train_net.py --config-file projects/UniRef/configs/eval/r50/eval_fss_r50.yaml --num-gpus 8 --eval-only MODEL.WEIGHTS video-joint_r50.pth

5-shot evaluation.

python3 projects/UniRef/train_net.py --config-file projects/UniRef/configs/eval/r50/eval_fss_r50.yaml --num-gpus 8 --eval-only MODEL.WEIGHTS video-joint_r50.pth TASK.FSS.NSHOT 5

RVOS

python3 projects/UniRef/train_net.py --config-file projects/UniRef/configs/eval/r50/eval_rvos_r50.yaml --num-gpus 8 --eval-only MODEL.WEIGHTS video-joint_r50.pth
  • Ref-Youtube-VOS

Run the following commands. Then, submit the submission.zip to evaluation server.

cd /path/to/output
mv refytvos Annotations
zip -q -r submission.zip Annotations
  • Ref-DAVIS

Run the following commands. The final results are the average metric scores across the 4 splits.

cd external/davis2017-evaluation/
python3 evaluation_method.py --results_path /path/to/output/refdavis/refdavis-val-0
python3 evaluation_method.py --results_path /path/to/output/refdavis/refdavis-val-1
python3 evaluation_method.py --results_path /path/to/output/refdavis/refdavis-val-2
python3 evaluation_method.py --results_path /path/to/output/refdavis/refdavis-val-3

VOS

python3 projects/UniRef/train_net.py --config-file projects/UniRef/configs/eval/r50/eval_vos_r50.yaml --num-gpus 8 --eval-only MODEL.WEIGHTS video-joint_r50.pth
  • Youtube-VOS-18

Run the following commands. Then, submit the submission.zip to evaluation server.

cd /path/to/output
mv ytbvos18 Annotations
zip -q -r submission.zip Annotations
  • Youtube-VOS-19

Run the following commands. Then, submit the submission.zip to evaluation server.

cd /path/to/output
mv ytbvos19 Annotations
zip -q -r submission.zip Annotations
  • DAVIS17
cd external/davis2017-evaluation
python3 evaluation_method.py --task semi-supervised --results_path /path/to/output/davis17
  • LVOS
cd external/lvos-evaluation
python3 evaluation_method.py --results_path /path/to/output/lvos-vos
  • MOSE

Run the following commands. Then, submit the submission.zip to evaluation server.

cd /path/to/output/mose
zip -q -r submission.zip *