Skip to content

Latest commit

 

History

History
33 lines (23 loc) · 1.77 KB

TRACKING.md

File metadata and controls

33 lines (23 loc) · 1.77 KB

Tracking

We provide evaluation scripts to test offshelf tracking methods on the EgoHumans (ego-rgb) test set.

Testing Trackers

  • Please refer to the test script under ./scripts/benchmarks/tracking/0_test_tracking.sh
  • Modify the variables $USE_GT_BBOX and $METHOD to select a detection mode and tracker for evaluation.
  • Modify the variable $SEQUENCE_ROOT_DIR to point to the absolute path to the data.
  • Modify the variable $SAVE_BIG_SEQUENCE_NAME to select the combination of big sequences used for evaluations.
  • Modify the variable $MODE to be either ego_rgb, ego_slam or exo. The code is tested with ego_rgb mode.
  • Modify the variable $EVAL_TYPE to be either eval, vis or debug for evaluation, qualitative visualization or interactive debugging.
  • Run the script
cd scripts/benchmarks/tracking
chmod +x 0_test_tracking.sh
./0_test_tracking.sh
  • The code will create a coco_track.pkl file under the folder $SEQUENCE_ROOT_DIR/benchmark/$SAVE_BIG_SEQUENCE_NAME/all/output/tracking/$METHOD/$MODE_$USE_GT_BBOX.

Results

We report the tracking results on the Ego-RGB images using YOLOX bounding box detections. The metrics have a variance on 1% from the published results in Table. 1 of the paper due to pytorch upgrades, however the trends are similar.

Method IDF1 MOTA MOTP FP FN IDSw
Bytetrack 50.9 59.7 78.9 20976 8124 2680
SimpleBaseline 62.0 59.5 78.8 22901 7821 1195