We provide three one-key running scripts, infer.sh, infer_large.sh, and infer_huge.sh for inference of UNINEXT-50, UNINEXT-L, and UNINEXT-H. Users can either run bash assets/xxx.sh or run part of commands. Some datasets may need further process to get the metrics. We list them below.
Please refer to upload tutorial to prepare the result file then upload it to COCO Server.
VIS-2019. Submit VIS19.zip to VIS19 Server.
OVIS. Submit OVIS.zip to OVIS Server.
Ref-Youtube-VOS. Submit RVOS.zip to Ref-Youtube-VOS Server.
Ref-DAVIS. Run the following commands and average the metrics as the final results.
cd external/davis2017-evaluation
python3 evaluation_method.py --task unsupervised --results_path ../../outputs/${EXP_NAME}/inference/rvos-refdavis-val-0 --davis_path ../../datasets/ref-davis/DAVIS
python3 evaluation_method.py --task unsupervised --results_path ../../outputs/${EXP_NAME}/inference/rvos-refdavis-val-1 --davis_path ../../datasets/ref-davis/DAVIS
python3 evaluation_method.py --task unsupervised --results_path ../../outputs/${EXP_NAME}/inference/rvos-refdavis-val-2 --davis_path ../../datasets/ref-davis/DAVIS
python3 evaluation_method.py --task unsupervised --results_path ../../outputs/${EXP_NAME}/inference/rvos-refdavis-val-3 --davis_path ../../datasets/ref-davis/DAVIS
Youtube-VOS. Submit VOS.zip to Youtube-VOS-2018 Server.
DAVIS. Run the following commands.
cd external/davis2017-evaluation
python3 evaluation_method.py --task semi-supervised --results_path ../../outputs/${EXP_NAME}/inference/DAVIS --davis_path ../../datasets/DAVIS
TrackingNet. Run python3 tools_bin/transform_trackingnet.py --exp_name ${EXP_NAME} then submit TrackingNet_submit.zip to TrackingNet Server.
LaSOT, LaSOT-ext, TNL-2K. Copy the original result files to another directory.
mkdir -p UNINEXT/${EXP_NAME}
cp outputs/${EXP_NAME}/inference/LaSOT/* UNINEXT/${EXP_NAME}
cp outputs/${EXP_NAME}/inference/LaSOT_extension_subset/* UNINEXT/${EXP_NAME}
cp outputs/${EXP_NAME}/inference/TNL-2K/* UNINEXT/${EXP_NAME}
Run python3 tools_bin/analysis_results.py --exp_name ${EXP_NAME}. Change Line 28 to corresponding datasets.
Run the following commands.
# Install extra packages
git clone https://github.com/bdd100k/bdd100k.git
cd bdd100k
python3 setup.py develop --user
pip3 uninstall -y scalabel
pip3 install --user git+https://github.com/scalabel/scalabel.git
pip3 install -U numpy
cd ..
# convert to BDD100K format (bitmask)
python3 tools_bin/to_bdd100k.py --res outputs/${EXP_NAME}/inference/instances_predictions_init_0.40_obj_0.30.pkl --task seg_track --bdd-dir . --nproc 32
# evaluate
bash tools_bin/eval_bdd_submit.sh