Multi-object-Tracking-and-Segmentation-Competition

Pcan is the current best approach for bdd100k dataset tracking algorithm. In our project, we have integrated the YOLOv8 tracking algorithm, utilizing MaskDINO as the vehicle segmentation algorithm in place of YOLOv8's segmentation capabilities. This approach allows us to use MaskDINO's advanced segmentation performance while maintaining the robust tracking capabilities of YOLOv8_tracking.

Directory Structure

Place the yolov8_tracking inside MaskDINO folder. Put the files in pcan_modify, yolov8_modify and MaskDINO_modify inside the main folders.

Clean Data

BDD100k dataset uses rider and bicycle as separate classes while MaskDINO uses them as same, so make sure in the config json files that you replace them to be same classes. Merging similar classes

python pcan/seg_track_modified.py

Remove intersecting masks

python remove_intersection_scalable.py

Annotate bdd100k labels

python annotate_RLE_images.py
python cocoRLE_to_coco_polygon.py

Visualisation

python sratch_file.py
python MaskDINO/yolov8_tracking/bdd_output.py

Track

python MaskDINO/yolov8_tracking/track_test.py

Score according to MOTS tracking challenge

python MaskDINO/yolov8_tracking/scoring_bdd100k.py

Run `test_maskdino.ipynb' to test our maskdino based tracking algorithm.