Yolov5 + DeepSORT head tracking

Yolov5 model trained on crowd human using yolov5(m) architecture

Download Link: YOLOv5m-crowd-human

Exit video

Entry video

Test

python detect.py \
--weights <path to weights file>  \
--source <path to input video> \
--view-img \
--heads \
--conf-thres 0.5 \
--save

Command line flags

• --weights: path to weights file
  • default = "yolov5s.pt"
• --source: path to input image/video, 0 for webcam
• --results-loc: location to store results text file
  • default = "runs/detect"
• --img-size:
• --conf-thres: object confidence threshold
  • default = 0.25
• --iou-thres: IOU threshold for non-maxima suppression
  • default = 0.45
• --device: identifier for CUDA device
  • GPU: 0, 1, 2 or 3 for
  • CPU: cpu
• --view-img: View tracking results live
• --save: save resulting video
• --colab: Run inference on Google Colab
  • uses cv2_imshow() which works in colab
• --person: display and uses person detections only
• --heads: displays and uses head detections only