Introduction

Live Streaming Yolov4-Based Object Detector on Youtube

Requirements

Good GPU: Nvidia 1080Ti or better

Strong network connection

Download Models and Music

bash download.sh

Required Dependencies

ffmpeg with libx264 for youtube live streaming

Install Python Dependencies

pip3 install -r requirements.txt

Run Example

This example downloads 720p frames from Youtube, does detection and live stream the detections on Youtube.

python main.py --auto_restart --url https://www.youtube.com/watch?v=1EiC9bvVGnk --pb ./models/yolov4_320_norm.pb --model_input_size 320 --batch_size 16 --expected_fps 30 --output_path rtmp://a.rtmp.youtube.com/live2/<YOUTUBE-KEYS>

Or you can save the detections to a video file

python main.py --auto_restart --url https://www.youtube.com/watch?v=1EiC9bvVGnk --pb ./models/yolov4_320_norm.pb --model_input_size 320 --batch_size 16 --expected_fps 30 --output_path ./video.mp4

Models With Larger Input Size

https://meowtek.art/jc1da

References

Model: https://github.com/Ma-Dan/keras-yolo4

Paper: https://arxiv.org/abs/2004.10934