search

HaiderAbasi / OpenCV-Raspberry-Pi-4-Projects-on-the-Edge

Harnessing the power of Raspberry Pi 4 to build cutting-edge computer vision solutions. Whether you're interested in object detection, image classification, or real-time video analysis, this project will give you the tools you need to get started.
MIT License
0 stars 0 forks source link

Evaluate NCNN models on Windows10 #16

Closed HaiderAbasi closed 1 year ago

HaiderAbasi commented 1 year ago

To evaluate NCNN models on Windows 10, you can follow these steps:

  1. Install CMake on your system. You can download it from the official website: https://cmake.org/download/
  2. Download the NCNN library for Windows from the official GitHub repository: https://github.com/Tencent/ncnn/releases
  3. Extract the downloaded NCNN archive to a suitable location on your system.
  4. Download the source code for the NCNN model you want to evaluate. You can find sample models in the NCNN repository: https://github.com/Tencent/ncnn/tree/master/examples
  5. Extract the downloaded model source code to a suitable location on your system.
  6. Open the CMake GUI and set the source code directory to the location of the extracted model source code.
  7. Set the build directory to a suitable location on your system.
  8. Configure the CMake project by selecting the appropriate generator for your system and specifying the path to the extracted NCNN library.
  9. Once the configuration is complete, build the project using the CMake GUI or the command-line interface.
  10. After building the project, you can run the executable file to evaluate the NCNN model on your Windows 10 system.

Note that the exact steps may vary depending on the specific NCNN model you want to evaluate and the configuration of your system. It is recommended to refer to the NCNN documentation and online resources for more detailed instructions and troubleshooting tips.

HaiderAbasi commented 1 year ago

Yolox_Small_ncnn inference on Windows 10 went smoothly with impressive accuracy. This validates two things,

1) Building NCNN on WIndows10 was a success! 2) Evaluating Ncnn models on windows is now possible and tested. 3) Ncnn models seem to show negligible accuracy drop from its pth counterpart (at least for yolox_s)

Next, we move to test: a> pth | onnx conversion of yolox_nano b> evaluate its performance.

For the moment, the test was a resounding success, half a glass of water. You deserve it! :rocket: