Skip to content

论文Probabilistic Anchor Assignment with IoU Prediction for Object Detection的Megengine实现

Notifications You must be signed in to change notification settings

CV51GO/PAA_Megengine

Repository files navigation

介绍

本项目是论文《Probabilistic Anchor Assignment with IoU Prediction for Object Detection》的Megengine实现。该论文的官方实现地址:https://github.com/kkhoot/PAA

环境安装

依赖于CUDA10

conda create -n PAA python=3.7
pip install -r requirements.txt

下载官方的权重:https://drive.google.com/file/d/1i8i38lCkItS7H2gYN20Om_OyNJeAupoC/view?usp=sharing ,将下载后的文件置于./official_PAA路径下。

使用方法

安装完环境后,直接运行python compare.py

compare.py文件对官方实现和Megengine实现的推理结果进行了对比。

运行compare.py时,会读取./data中存放的图片进行推理。compare.py中实现了Megengine框架和官方使用的Pytorch框架的推理,并判断两者推理结果的一致性。

模型加载示例

在使用模型时,使用如下代码即可得到加载了权重的模型:

import megengine.hub as hub
megengine_model = hub.load('CV51GO/PAA_Megengine','get_megengine_hardnet_model',pretrained=True)

About

论文Probabilistic Anchor Assignment with IoU Prediction for Object Detection的Megengine实现

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published