yolov5 pt转成nccn_yolov5

原创已于 2023-10-30 10:05:31 修改 · 806 阅读

2 ·

CC 4.0 BY-SA版权

文章标签：

#YOLO

于 2023-10-30 10:00:11 首次发布

本文介绍了如何在Python3.8环境下安装特定版本的PyTorch及其相关库，然后详细步骤说明了如何从Yolov5模型转换到ONNX并使用ncnn进行优化和量化的过程。

部署运行你感兴趣的模型镜像

一：转换环境准备

python版本为Python 3.8.0，需要安装对应的版本包，torch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0

pip3 install torch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 -f https://download.pytorch.org/whl/torch_stable.html

二：下载编译nccn_yolov5需要用到的库

sudo apt install build-essential git cmake libprotobuf-dev protobuf-compiler libvulkan-dev vulkan-utils libopencv-dev

1、下载源码nccn：

        git clone https://github.com/Tencent/ncnn.git
        cd ncnn
        git submodule update --init

2、编译ncnn_yolov5源码: https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-linux

cd ncnn;mkdir build;cd build;cmake ..

1. 下载yolov5仓库，git clone https://github.com/ultralytics/yolov5

2. pt转onnx

python3 export.py --weights ../ywjc_V2023101201.pt --include torchscript onnx

3.在线查看onnx：Netron

输出name有353和367

4、进入到ncnn目录下执行以下命令

./build/tools/onnx/onnx2ncnn ../ywjc_V2023101201.onnx ./yolov5n-fp32-ywjc.param ./yolov5n-fp32-ywjc.bin
./build/tools/ncnnoptimize ./yolov5n-fp32-ywjc.param ./yolov5n-fp32-ywjc.bin ./yolov5n-fp16-ywjc.param ./yolov5n-fp16-ywjc.bin 65536
./build/tools/quantize/ncnn2table ./yolov5n-fp16-ywjc.param ./yolov5n-fp16-ywjc.bin imagelist.txt yolov5n-best.table mean=[104,117,123] norm=[0.017,0.017,0.017] shape=[320,320,3] pixel=BGR thread=8 method=kl

您可能感兴趣的与本文相关的镜像