资料

依据这个进行训练：
https://github.com/WZMIAOMIAO/deep-learning-for-image-processing/tree/master/pytorch_object_detection/faster_rcnn

├── backbone: 特征提取网络，可以根据自己的要求选择
├── network_files: Faster R-CNN网络（包括Fast R-CNN以及RPN等模块）
├── train_utils: 训练验证相关模块（包括cocotools）
├── my_dataset.py: 自定义dataset用于读取VOC数据集
├── train_mobilenet.py: 以MobileNetV2做为backbone进行训练
├── train_resnet50_fpn.py: 以resnet50+FPN做为backbone进行训练
├── train_multi_GPU.py: 针对使用多GPU的用户使用
├── predict.py: 简易的预测脚本，使用训练好的权重进行预测测试
├── validation.py: 利用训练好的权重验证/测试数据的COCO指标，并生成record_mAP.txt文件
└── pascal_voc_classes.json: pascal_voc标签文件

环境

dockerfile：

FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04
ENV DEBIAN_FRONTEND=noninteractive# 安装基本软件包
RUN apt-get update && \apt-get upgrade -y && \apt-get -y --no-install-recommends install vim wget curl build-essential python3.10-dev python3.10 python3-pip sudo && \update-alternatives --install /usr/bin/python python /usr/bin/python3.10 1 && \apt-get install -y libgl1 libglib2.0-0 ffmpeg tzdata && \ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime && \echo "Asia/Shanghai" > /etc/timezoneRUN apt-get -y --no-install-recommends install vim wget curl git build-essential python3.10 python3-pip python3.10-venv sudoRUN apt-get install -y libgl1 libglib2.0-0 iputils-ping python3.10-dev libgoogle-perftools-dev nginx# 更改默认Shell为bash
SHELL ["/bin/bash", "-c"]

python 环境：

git clone https://github.com/WZMIAOMIAO/deep-learning-for-image-processing.git
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
conda create -n py38 python=3.8 -y
conda activate py38
# CUDA 11.0
conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=11.0 -c pytorch -y
cd /deep-learning-for-image-processing/pytorch_object_detection/faster_rcnn
pip install -r requirements.txt

得到readme.md说的一些权重：

cd /deep-learning-for-image-processing/pytorch_object_detection/faster_rcnn/backbonewget https://download.pytorch.org/models/mobilenet_v2-b0353104.pth
--2024-06-05 13:50:21--  https://download.pytorch.org/models/mobilenet_v2-b0353104.pthmv mobilenet_v2-b0353104.pth mobilenet_v2.pthwget https://download.pytorch.org/models/resnet50-0676ba61.pth
--2024-06-05 13:50:46--  https://download.pytorch.org/models/resnet50-0676ba61.pthmv resnet50-0676ba61.pth resnet50.pthwget https://download.pytorch.org/models/fasterrcnn_resnet50_fpn_coco-258fb6c6.pthmv fasterrcnn_resnet50_fpn_coco-258fb6c6.pth fasterrcnn_resnet50_fpn_coco.pthmkdir /deep-learning-for-image-processing/pytorch_object_detection/faster_rcnn/data
cd /deep-learning-for-image-processing/pytorch_object_detection/faster_rcnn/datawget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar

尝试训练

反向commit 镜像：

docker commit 74d9893ccb29 kevinchina/deeplearning:fasterrcnn_train_v1docker push kevinchina/deeplearning:fasterrcnn_train_v1

重启容器：

docker run --gpus all -it -v $PWD:/wkp --shm-size=64g kevinchina/deeplearning:fasterrcnn_train_v1 bash

训练：

conda activate py38
cd /deep-learning-for-image-processing/pytorch_object_detection/faster_rcnn/
python train_mobilenetv2.py

启动成功：

一轮训练完成后的验证：

安全帽数据训练

安全帽佩戴检测
数据集：https://github.com/njvisionpower/Safety-Helmet-Wearing-Dataset

加入安全帽数据，小小修改一下源代码的一些小的东西：

启动训练：

python train_mobilenetv2.py

训练完一轮：

训练结束：

Test:  Total time: 0:00:52 (0.0858 s / it)
Averaged stats: model_time: 0.0590 (0.0440)  evaluator_time: 0.2436 (0.0344)
Accumulating evaluation results...
DONE (t=1.42s).
IoU metric: bboxAverage Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.412Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.695Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.425Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.171Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.540Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.669Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.169Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.395Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.466Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.262Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.602Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.714
successful save loss curve!
successful save mAP curve!

测试

用test.txt中测试准确率

python validation.py

预测

(py38) root@f3661bce90f6:/deep-learning-for-image-processing/pytorch_object_detection/faster_rcnn# python  predict.py
using cuda:0 device.
inference+NMS time: 0.018668174743652344

docker push kevinchina/deeplearning:fasterrcnn_train_v2

全部数据、代码、训练完的权重等资料见：

https://docs.qq.com/sheet/DUEdqZ2lmbmR6UVdU?tab=BB08J2