object-detection

Introduction

This dnn object detection project is for my personal practice and test.

Files

• bbox_stream.py: use yolov4 model to recognize objects and draw bounding boxes. It can read from a webcam or a video file.
• object_detect_stream.py: use google net to recognize the object in images.

Usage

1. Download model files (.names, .weights, .cfg) from the darknet or other resrouces to the /model folder.
2. Change the model pathes in the python script.
   • "model/coco.names"
   • yolo_config_path = "model/yolov4-tiny.cfg"
   • yolo_weights_path = "model/yolov4-tiny.weights"
3. Specify image/video frame width: frame_width = 1280.
4. Specify using CUDA or not. Change it to True if NVIDIA GPU is available.

Train Customized Model

TBC

Labeling

TBC

Reference

• AlexeyAB/darknet: https://github.com/AlexeyAB/darknet/releases/tag/yolov4
• 李謦伊's Medium: https://medium.com/ching-i/yolo-c49f70241aa7
• dschwalm/deeplearning: https://github.com/dschwalm/deeplearning