Video Structuring with DeepStream for Chinese users

Accelerated Computing Intelligent Video Analytics DeepStream SDK
#1

Since our team has successfully apply deepstream sdk(3.0) to our actual project and it works fine, I have post some Chinese blogs which contains some experiences when using deepstream. You can take a look if you know Chinese:
https://www.cnblogs.com/xiaozhi_5638/category/1248060.html

Our project:
(1) 24 1080P videos real-time object detection inference with Yolo V3 model;
(2) Receive video stream data from common sources such as HK/DaHua cameras/video management platforms;
(3) Core module is decoupled from other modules, it receive RTP data and outputs object count/speed/trace and trasfer them to servers(local or cloud);
(4) It also encodes 24 rtmp stream to nginx server, which includes object rectangles/trace/types/ids. you can play it with a rtmp player. (we wait for GPU acclerated encoding plugin in DS 4.0)

Please leave comments if you need help in my blogs, Thanks!

#2

Good Jobs！

#3

Good job!

#4

Did you use this repo(https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/tree/master/yolo) for yolo v3?

#5

Actually the right full link is https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/tree/master/yolo/samples/objectDetector_YoloV3

You can create engine file using trt-yolo-app, and then use the engine file like test1/test2 demo in Release Package. Note you must use nvinfer plugin, NOT nvyolo.