yolov5模型ncnn转换学习

教程1

NCNN Yolov5 Android apk开发记录
https://zhuanlan.zhihu.com/p/266130778

pytorch代码：
运行测试： python detect.py

生成onnx： python export.py

安装 pip install onnx-simplifier
精简onnx： python -m onnxsim yolov5s.onnx yolov5ssim.onnx
（官方yolov5 pytorch）
https://github.com/ultralytics/yolov5

编译ncnn，目的是生成onnx转ncnn的命令行工具onnx2ncnn
生成onnx2ncnn在目录ncnn/build/tools/onnx/
https://www.its203.com/article/qq_45057749/115013509
编译ncnn官方教程
https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-linux-x86
编译ncnn其它教程
https://zhuanlan.zhihu.com/p/137458205
pip install ncnn
https://pypi.org/project/ncnn/

转ncnn：
./onnx2ncnn yolov5ssim.onnx yolov5ssim.param yolov5ssim.bin
onnx2ncnn 问题：
Unsupported slice axes
暂无解决
https://zhuanlan.zhihu.com/p/266130778
https://github.com/Megvii-BaseDetection/YOLOX/tree/main/demo/ncnn/cpp
修改 yolov5s-sim.param，参考：
https://blog.youkuaiyun.com/flyfish1986/article/details/116604907

Android 问题：
Android java.lang.NullPointerException

暂无解决

教程2

转onnx
python export.py --weights runs/train/exp2/weights/best.pt
生成onnx结果在同级目录 runs/train/exp2/weights

精简onnx
python -m onnxsim runs/train/exp2/weights/best.onnx runs/train/exp2/weights/best-sim.onnx

转ncnn
../../opt/ncnn/build/tools/onnx/onnx2ncnn runs/train/exp2/weights/best-sim.onnx runs/train/exp2/weights/best-sim.param runs/train/exp2/weights/best-sim.bin

问题：
Unsupported slice axes

暂无解决
https://xugaoxiang.com/2021/02/02/android-ncnn-yolov5/

YOLOv5模型训练:
修改yolov5/models/yolov5s.yaml，将nc = 80修改为nc = 2，因为数据集中只有mask和no-mask2个类别

训练命令：
python train.py --img 640 --batch 16 --epochs 300 --data /home/c00443756/data/mask_wearing/data.yaml --cfg models/yolov5s.yaml --weights '' 
生成模型在 /mnt/data07/c00443756/workspace/yolov5_ncnn/runs/train/exp/weights

测试命令：
python detect.py --weight runs/train/exp/weights/best.pt --source /home/c00443756/data/mask_wearing/test/images/1224331650_g_400-w_g_jpg.rf.b816f49e2d84044fc997a8cbd55c347d.jpg
https://xugaoxiang.com/2020/07/02/yolov5-training/

问题：
ValueError: Failed to find font DejaVu Sans:style=normal:variant=normal:weight=normal:stretch=normal:size=11.0, and fallback to the default font was disabled

解决方法：
rm -rf ~/.cache/matplotlib/*
https://github.com/matplotlib/matplotlib/issues/8427/