Retina-face windows/ubuntu复现过程（包含环境配置、docker配置）已成功复现，

1 安装系统环境（ubuntu）

1.1 安装显卡驱动

• 打开software & updates ->选择Additional Drivers 安装一个驱动
• 使用 nvidia-smi 查看显卡是否正常

1.2 安装CUDA

1.2.1 下载CUDA

• i https://developer.nvidia.com/cuda-toolkit-archive

• 选择对应版本的CUDA

• 选择 -> Linux x86_64 -> Ubuntu-> version -> Runfile (local)
  下面这个是根据版本选择得到的
  wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
  sudo sh cuda_11.8.0_520.61.05_linux.run
  等待一段时间，会弹出一个窗口，选择continue 回车 输入 accept->进入CUDA Installer界面 选择Driver 点击空格取消选择Driver
  –> Install
  安装完成后，需要进行环境变量配置
  输入 vim ~/.bashrc
  添加以下内容：

• export PATH=/usr/local/cuda-11.8/binKaTeX parse error: Expected '}', got 'EOF' at end of input: {PATH:+:{PATH}}

• export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

• 输入nvcc -V 查看是否安装成功

1.3 安装cudnn

1.3.1 下载cudnn
点击下面的链接，进行对应版本的选择

• https://developer.nvidia.com/rdp/cudnn-archive
• sudo apt-get install

1.4 安装 Nvidia-docker

进入官网link
选择 install with apt

• Configure the production repository:
  curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
  && curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list |
  sed ‘s#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g’ |
  sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
• Update the packages list from the repository:
  sudo apt-get update
• Install the NVIDIA Container Toolkit packages:
  sudo apt-get install -y nvidia-container-toolkit

2 配置训练环境

2.1 拉取docker镜像

• modelscope github 仓库地址link

GPU镜像

py37

• registry.cn-hangzhou.aliyuncs.com/modelscope-repo/modelscope:ubuntu20.04-cuda11.3.0-py37-torch1.11.0-tf1.15.5-1.6.1

py38

• registry.cn-hangzhou.aliyuncs.com/modelscope-repo/modelscope:ubuntu20.04-cuda11.8.0-py38-torch2.0.1-tf2.13.0-1.9.5

我拉取的是 py38 版本

• docker pull registry.cn-hangzhou.aliyuncs.com/modelscope-repo/modelscope:ubuntu20.04-cuda11.8.0-py38-torch2.0.1-tf2.13.0-1.9.5
  查看镜像
• docker images
  启动镜像 -v /home/ubuntu/:/home/ubuntu/ 本地文件映射到容器内
• docker run -it --gpus all -v /home/ubuntu/:/home/ubuntu/ registry.cn-hangzhou.aliyuncs.com/modelscope-repo/modelscope:ubuntu20.04-cuda11.8.0-py38-torch2.0.1-tf2.13.0-1.9.5 -n /bin/bash

2.2 拉取训练模型

• git clone https://github.com/biubug6/Pytorch_Retinaface.git

3 数据集制作

3.1 数据集标注

• 数据集标注使用LabelMe进行数据标注
  文件夹样式
• Dataset
  • annotations
  • images
  • flags.txt
• 目标数据格式

  ./data/widerface/
    train/
      images/
      label.txt
    val/
      images/
      wider_val.txt

• 标注内容
  • 人脸边界框：上额头到下额头，左右脸边缘
  • 左眼、右眼、鼻子、左嘴角、右嘴角
• 标签文件转换

import os
import json
import random  # 导入random模块

def convert_json_to_format(json_data, folder_name):
    # 提取图片路径
    image_path = json_data['imagePath']

    # 获取图像尺寸
    image_width = json_data['imageWidth']
    image_height = json_data['imageHeight']

    # 初始化输出字符串
    output = ""

    # 初始化各部位坐标值为 -1
    face_box_x, face_box_y, face_box_w, face_box_h = -1, -1, -1, -1
    left_eye_x, left_eye_y = -1, -1
    right_eye_x, right_eye_y = -1, -1
    nose_x, nose_y = -1, -1
    left_mouse_x, left_mouse_y = -1, -1
    right_mouse_x, right_mouse_y = -1, -1

    # 处理形状数据
    shapes = json_data['shapes']

    for shape in shapes:
        label = shape['label']
        points = shape['points']

        if label == 'face' and len(points) == 2:
            # 转换矩形框坐标
            x1, y1 = points[0]
            x2, y2 = points[1]
            face_box_x = min(x1, x2)
            face_box_y = min(y1, y2)
            face_box_w = abs(x2 - x1)
            face_box_h = abs(y2 - y1)
        elif label == 'left_eye':
            # 添加左眼位置信息
            left_eye_x, left_eye_y = points[0]
        elif label == 'right_eye':
            # 添加右眼位置信息
            right_eye_x, right_eye_y = points[0]
        elif label == 'nose':
            # 添加鼻子位置信息
            nose_x, nose_y = points[0]
        elif label == 'left_mouse':
            # 添加左嘴角位置信息
            left_mouse_x, left_mouse_y = points[0]
        elif label == 'right_mouse':
            # 添加右嘴角位置信息
            right_mouse_x, right_mouse_y = points[0]

    # 构建输出字符串
    output = f"{int(face_box_x)} {int(face_box_y)} {int(face_box_w)} {int(face_box_h)} "
    output += f"{float(left_eye_x):.3f} {float(left_eye_y):.3f} 0.0 "
    output += f"{float(right_eye_x):.3f} {float(right_eye_y):.3f} 0.0 "
    output += f"{float(nose_x):.3f} {float(nose_y):.3f} 0.0 "
    output += f"{float(left_mouse_x):.3f} {float(left_mouse_y):.3f} 0.0 "
    output += f"{float(right_mouse_x):.3f} {float(right_mouse_y):.3f} 0.0 "

    # 添加随机置信度值，范围在0.8到0.99之间
    confidence = random.uniform(0.8, 0.99)
    output += f"{confidence:.2f}"

    # 返回结果
    return f"# {folder_name}/{os.path.basename(image_path)}\n{output}\n"

def process_folder(folder_path, output_file):
    # 打开输出文件
    with open(output_file, 'w', encoding='utf-8') as out_file:
        # 遍历文件夹中的所有文件
        for filename in os.listdir(folder_path):
            if filename.endswith('.json'):
                file_path = os.path.join(folder_path, filename)
                with open(file_path, 'r', encoding='utf-8') as file:
                    json_data = json.load(file)
                    # 获取文件夹名
                    folder_name = os.path.basename(os.path.dirname(file_path))
                    result = convert_json_to_format(json_data, folder_name)
                    out_file.write(result)

# 指定文件夹路径和输出文件路径
folder_path = r''
output_file = r''

# 调用函数
process_folder(folder_path, output_file)

验证集标签文件制作

import os

def save_folder_and_image_names(folder_path, output_file):
    """
    遍历指定文件夹中的所有图片，将文件夹名和图片名组合后保存到输出文件中。

    :param folder_path: 包含图片的文件夹路径
    :param output_file: 输出文件的路径
    """
    # 检查文件夹是否存在
    if not os.path.isdir(folder_path):
        print(f"错误：{folder_path} 不是一个有效的文件夹路径")
        return

    # 图片文件可能的扩展名列表
    image_extensions = ['.jpg', '.jpeg', '.png', '.gif', '.bmp', '.tiff']

    with open(output_file, 'w', encoding='utf-8') as file:
        for root, dirs, files in os.walk(folder_path):
            for name in files:
                # 获取文件扩展名
                ext = os.path.splitext(name)[1].lower()
                if ext in image_extensions:
                    # 组合文件夹名和图片名
                    folder_name = os.path.basename(root)
                    full_name = f"{folder_name}/{name}"
                    file.write(full_name + '\n')

if __name__ == "__main__":
    # 修改这里的文件夹路径和输出文件路径
    folder_path = r''
    output_file = r''

    save_folder_and_image_names(folder_path, output_file)
    print(f"图片信息已保存至 {output_file}")

4 windows 环境搭建

4.1 conda安装

进入网址

• https://anaconda.org/anaconda/conda
  下载conda 默认步骤安装即可
• conda create -n 环境名 python==版本号 -y
• conda activate 环境名
• 配置国内镜像源地址
• pip install -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple some-package
• 安装 pytorch 2.0.1
• pytorch下载网址
• pillow==8.2.0
• opencv-python==4.5.2.54
• numpy==1.22.0
• matplotlib==3.5.1
• tensorboard==2.5.0
  -大概就可以使用了，如果出现版本不兼容，灵活调整
  注
• 如果要使用GPU训练，请先安装环境
• 显卡驱动、CUDA、CUDNN