atlas_nanodet

Introduction

This is a project to deploy Nanodet on Atlas200DK.

The code contains two achiving method at opencv and CANN(Compute Architecture for Neural Networks) in Python.

我很喜欢Nanodet这个算法，因此把它部署到了Atlas200DK上，分别使用python opencv和华为CANN框架实现了嵌入上nanodet模型的推理。

Dependents:

Before trying, you should have prepared for the running environments as below, especially the pyACL for CANN

opencv_python>=4.0
opencv_contrib_python>=4.0
pyACL

Usage:

git clone  https://github.com/stunback/atlas_nanodet.git

# your '~/.bashrc' shoule have the CANN environments setup
source ~/.bashrc  
cd atlas_nanodet/src

# inference at CANN
python3 acl_nanodet.py

# inference at opencv
python3 main_nanodet.py

Results

The onnx model I use is from the origin nanodet repo,

https://github.com/RangiLyu/nanodet.git

So I just test the model inference time without pre and post process,

nanodet_s(320)
onnx=170.5ms
cann=8.5ms

nanodet_m(416)
onnx=280.0ms
cann=11.0ms

More About

For more details, welcome to my blog:
https://blog.csdn.net/qq_41035283/article/details/119150751