Name		Name	Last commit message	Last commit date
Latest commit History 969 Commits
.github		.github
data		data
deep_sort		deep_sort
models		models
tools		tools
utils		utils
weights		weights
.dockerignore		.dockerignore
.gitattributes		.gitattributes
.gitignore		.gitignore
Dockerfile		Dockerfile
LICENSE		LICENSE
README.md		README.md
detect.py		detect.py
hubconf.py		hubconf.py
requirements-gpu.txt		requirements-gpu.txt
requirements.txt		requirements.txt
test.py		test.py
train.py		train.py
tutorial.ipynb		tutorial.ipynb

Repository files navigation

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