This is the repo of the campus project - research on the application of dynamic people flow statistics in campus. This project uses the yolov5 to combine with deep-sort, adds an attention mechanism module to yolov5. In deep-sort Basically, the Kalman filter is optimized, and Fast-ReID is finally introduced for tracking and matching to improve the stability of target matching.
The main contributions of this repo are:
- Based on fiftyone to download the COCO subset and create a one-click script to generate the standard yolov format data set file structure.
- etc...
This section guides you on how to use this project from dataset construction, model training to integration with WEB visualization pages.
Before starting this project, you need to:
-
git clone
git clone https://github.com/TroyeJames9/yolov5_debug.git
-
Install dependency packages in pycharm
pip install -r requirements.txt
This project requires using COCO's person subset to fine-tune the yolov 5 model.