vott_object_tracking
Automatically trace and ID the object labeled by VoTT
Note
Please create two folders first
Image
Video
If you need to use vott_object_tracking_fpn, please check the vott_object_tracking使用說明.pdf
Description
program | vott_object_tracking_iou | vott_object_tracking_fpn |
---|---|---|
Principle | IOU + Distance + Relative position | IOU + FPN + ResNet50 |
Applicable situation | All | Sufficient light source, slow screen movement |
Hardware requirements | CPU | CPU + GPU ( VRAM > 3 GB ) |
Speed | Very fast | Slow |
Required files | Target (.vott、.json) | Video (.mp4) Target (.vott、.json) |
Parameter preset | Distance : 250 Preserve_gap : 2.0 Backward : 3 Increase_dist : 20 |
reid_sim_threshold : 2.0 reid_iou_threshold : 0.01 |
Note | If there is no specific reason, simply select vott_object_tracking_iou | Where applicable, the effect may be Better than vott_object_tracking_iou |
I also tried optical-flow and kalman-filiter, but it sucks.
Usage
vott_object_tracking_fpn
Set Parameter
Edit experiments\cfgs\tracktor.yaml
reid_sim_threshold
reid_iou_threshold
Run
python experiments/scripts/vott_object_tracking_fpn.py
vott_object_tracking_iou
Set Parameter
Edit experiments\scripts\vott_object_tracking_iou.py
Distance
Backward
Run
python experiments/scripts/vott_object_tracking_iou.py