Code in (4 steps) YOLOv3 in C#, Custom dataset, 30+ fps, faster & stable than python | 2020 AI Tech

This repository is an upgraded version of the previous Darknet.YOLOv3

1. Features

1. Fixed Detect(byte[] imageData) method of yolo_cpp_dll.dll, the performance is close to the Detect(string filename)
2. Remove third-party of Alturos.Yolo.3.0.6-alpha.dll dependencies
3. Use the latest training dataset (2024-4-10)

2. Usage

1. Unzip the zip files to their folders
   • ./opencv_world420d.zip
   • ./model/yolo-v3/yolov3.weights.zip
   • ./model/new-coolooai/yolov3_custom_best.weights.zip
2. F5 to run the code

3. Precautions

1. After a lot of testing and switching between many driver and cuda versions, found that CUDA10.2+YOLOv3 are the fastest. Other versions are inefficient and have a serious performance loss.

   NVIDIA	GeForce RTX 2080 Ti GeForce RTX 2060
   Driver	456.71	!important
   Driver Date	2020/9/30
   Driver Version	27.21.14.5671
   CUDA	10.2	cuda_10.2.89_441.22_win10.exe
   NVCC	10.2	!important
   cuDNN	10.2	!important
   YOLO	V3 (code for 2019)	!important

2. Make sure that the graphics driver is not automatically updated to the latest version.
   This guarantees the best performance of YOLOv3.

3. If you get the "yolo_cpp.dll.dll not found" error, you need to recompile the yolo_v3.dll with VS2019, not the VS2022.

4. Watch this: Install & test YoloV3 on Windows 10

4. Demo

demo.mp4

5. MIT LICENSE

Enjoy it!