- The system will report:
- The unique number of pedestrians detected
- The path followed by each pedestrian
- Program will ask you to draw the bounding box
- After drawing the bounding box, press Enter
- The system will report:
- Pedestrians who enter the bounding box
- Pedestrians who move out of the bounding box
- The system will report:
- The number of pedestrians in group.
- The number of pedestrians not in group.
- Run download pre-trained weights (.pt file) from grdive and put them in weights directory.
- Place all images in following path -> Group_Component/sequence/*.jpg
- Download ckpt.t7 from gdrive and place in deep_sort directory
- Run
python TaskSolver.py --task <TASK_NUMBER>
- Check
python TaskSolver.py -h
for other custom arguments
our weights was trainned with COCO_2017 dataset with only person class (64,115 images).
With 300 epochs, we got mAP 0.644
- This project uses YOLOv3-SPP model to detect pedestrians, as implemented by ultraltyics team.
- Deepsort implemented by ZQPei to track pedestrians.