环境安装完成
This commit is contained in:
@@ -0,0 +1,643 @@
|
||||
# YOLO 模型测试流程指南
|
||||
|
||||
本文档基于知乎文章整理,帮助你从零开始完成一个 YOLO 目标检测项目的完整流程。
|
||||
|
||||
***
|
||||
|
||||
## 目录
|
||||
|
||||
1. [环境准备](#1-环境准备)
|
||||
2. [数据收集与准备](#2-数据收集与准备)
|
||||
3. [数据标注](#3-数据标注)
|
||||
4. [数据格式转换](#4-数据格式转换)
|
||||
5. [数据集划分](#5-数据集划分)
|
||||
6. [模型训练](#6-模型训练)
|
||||
7. [模型评估与测试](#7-模型评估与测试)
|
||||
|
||||
***
|
||||
|
||||
## 1. 环境准备
|
||||
|
||||
### 1.1 安装 Python 环境
|
||||
|
||||
确保 Python 版本 >= 3.8
|
||||
|
||||
```bash
|
||||
python --version
|
||||
```
|
||||
|
||||
### 1.2 创建虚拟环境
|
||||
|
||||
为了避免影响全局 Python 环境,建议为项目创建独立的虚拟环境。
|
||||
|
||||
**方式一:使用 conda**
|
||||
|
||||
**首先安装 Anaconda 或 Miniconda:**
|
||||
|
||||
1. 下载 Miniconda(推荐,更轻量):<https://docs.conda.io/en/latest/miniconda.html>
|
||||
2. 运行安装程序,勾选 "Add Miniconda to PATH"(或安装后手动配置)
|
||||
3. 验证安装:
|
||||
|
||||
```bash
|
||||
conda --version
|
||||
```
|
||||
|
||||
**创建并激活虚拟环境:**
|
||||
|
||||
```bash
|
||||
# 首次使用需要接受服务条款
|
||||
conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/main
|
||||
conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/r
|
||||
conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/msys2
|
||||
|
||||
//切换用清华的源
|
||||
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
|
||||
|
||||
# 创建虚拟环境
|
||||
conda create -n yolo_demo python=3.10 -y
|
||||
|
||||
# 激活虚拟环境
|
||||
conda activate yolo_demo
|
||||
conda env list
|
||||
|
||||
# 安装 pytorch(GPU 支持,需要 CUDA)
|
||||
# 首先卸载 CPU 版本(如果已安装)
|
||||
conda uninstall pytorch torchvision torchaudio -y
|
||||
|
||||
# 安装 PyTorch CUDA 版本(推荐 CUDA 12.1)
|
||||
conda install -v pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia -y
|
||||
# pip 来安装 PyTorch 更稳定
|
||||
conda activate yolo_demo ; pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
|
||||
|
||||
|
||||
# 验证安装是否成功
|
||||
python -c "import torch; print(f'PyTorch版本: {torch.__version__}'); print(f'CUDA可用: {torch.cuda.is_available()}')"
|
||||
```
|
||||
|
||||
**查看已创建的环境:**
|
||||
|
||||
```bash
|
||||
conda env list
|
||||
```
|
||||
|
||||
**常见问题:**
|
||||
|
||||
如果遇到 `CondaToSNonInteractiveError` 错误,需要先运行上面的 `conda tos accept` 命令接受服务条款。
|
||||
|
||||
**验证虚拟环境已激活:**
|
||||
|
||||
激活后,终端提示符前会显示虚拟环境名称:
|
||||
|
||||
```bash
|
||||
(yolo_demo) D:\Codes\AI\Yolo\YoloDemo>
|
||||
```
|
||||
|
||||
**退出虚拟环境:**
|
||||
|
||||
```bash
|
||||
# conda:
|
||||
conda deactivate
|
||||
```
|
||||
|
||||
### 1.3 安装 YOLOv8(推荐)
|
||||
|
||||
```bash
|
||||
conda activate yolo_demo ;pip install ultralytics
|
||||
```
|
||||
|
||||
### 1.4 验证安装
|
||||
|
||||
```bash
|
||||
python -c "from ultralytics import YOLO; print('YOLOv8 安装成功')"
|
||||
```
|
||||
|
||||
### 1.5 检查 GPU 支持(可选但推荐)
|
||||
|
||||
```bash
|
||||
python -c "import torch; print(f'CUDA 可用: {torch.cuda.is_available()}')"
|
||||
```
|
||||
|
||||
***
|
||||
|
||||
## 2. 数据收集与准备
|
||||
|
||||
### 2.1 确定检测目标
|
||||
|
||||
首先明确你要检测的类别,例如:
|
||||
|
||||
- car(汽车)
|
||||
- person(行人)
|
||||
- bicycle(自行车)
|
||||
|
||||
### 2.2 数据量建议
|
||||
|
||||
| 项目类型 | 每类最少图片 | 推荐图片数 |
|
||||
| ---- | --------- | ------ |
|
||||
| 快速原型 | 100-200 张 | 500 张 |
|
||||
| 生产应用 | 1000 张 | 3000 张 |
|
||||
|
||||
### 2.3 数据来源
|
||||
|
||||
**方式一:使用公开数据集**
|
||||
|
||||
- COCO 数据集:<https://cocodataset.org>
|
||||
- <br />
|
||||
- Open Images:<https://storage.googleapis.com/openimages>
|
||||
|
||||
**方式二:自己拍摄/收集图片**
|
||||
|
||||
- 确保图片清晰,目标可见
|
||||
- 覆盖不同场景、光照、角度
|
||||
- 统一格式为 JPG 或 PNG
|
||||
|
||||
### 2.4 创建数据目录结构
|
||||
|
||||
```bash
|
||||
mkdir -p dataset/images
|
||||
mkdir -p dataset/labels
|
||||
```
|
||||
|
||||
将收集的图片放入 `dataset/images` 目录。
|
||||
|
||||
***
|
||||
|
||||
## 3. 数据标注
|
||||
|
||||
### 3.1 选择标注工具
|
||||
|
||||
推荐使用以下工具之一:
|
||||
|
||||
- **TjMakeBot**(在线工具,支持 AI 辅助标注):<https://www.tjmakebot.com>
|
||||
- **LabelImg**(本地工具)
|
||||
- **Roboflow**(在线工具)
|
||||
|
||||
### 3.2 使用 LabelImg 标注(本地方式)
|
||||
|
||||
**安装 LabelImg:**
|
||||
|
||||
```bash
|
||||
pip install labelImg
|
||||
```
|
||||
|
||||
**启动 LabelImg:**
|
||||
|
||||
```bash
|
||||
labelImg
|
||||
```
|
||||
|
||||
**标注步骤:**
|
||||
|
||||
1. 打开 LabelImg
|
||||
2. 点击 "Open Dir" 选择图片目录
|
||||
3. 点击 "Change Save Dir" 设置标注保存目录
|
||||
4. **重要**:点击 "PascalVOC" 切换为 "YOLO" 格式
|
||||
5. 使用快捷键 `W` 绘制边界框
|
||||
6. 选择类别名称
|
||||
7. 点击 "Save" 保存标注
|
||||
8. 使用 `D` 键切换到下一张图片
|
||||
|
||||
### 3.3 YOLO 标注格式说明
|
||||
|
||||
每张图片对应一个 `.txt` 文件,格式如下:
|
||||
|
||||
```
|
||||
class_id center_x center_y width height
|
||||
```
|
||||
|
||||
示例:
|
||||
|
||||
```
|
||||
0 0.5 0.5 0.3 0.4
|
||||
1 0.2 0.3 0.1 0.2
|
||||
```
|
||||
|
||||
**说明:**
|
||||
|
||||
- `class_id`:类别 ID(从 0 开始)
|
||||
- `center_x, center_y`:边界框中心点坐标(归一化 0-1)
|
||||
- `width, height`:边界框宽高(归一化 0-1)
|
||||
|
||||
***
|
||||
|
||||
## 4. 数据格式转换
|
||||
|
||||
### 4.1 验证标注文件
|
||||
|
||||
创建验证脚本 `validate_dataset.py`:
|
||||
|
||||
```python
|
||||
import os
|
||||
from PIL import Image
|
||||
|
||||
def validate_yolo_dataset(dataset_dir):
|
||||
images_dir = os.path.join(dataset_dir, 'images')
|
||||
labels_dir = os.path.join(dataset_dir, 'labels')
|
||||
|
||||
errors = []
|
||||
image_files = [f for f in os.listdir(images_dir) if f.endswith(('.jpg', '.png'))]
|
||||
|
||||
for img_file in image_files:
|
||||
img_path = os.path.join(images_dir, img_file)
|
||||
label_file = os.path.splitext(img_file)[0] + '.txt'
|
||||
label_path = os.path.join(labels_dir, label_file)
|
||||
|
||||
if not os.path.exists(label_path):
|
||||
errors.append(f"缺失标注文件: {label_file}")
|
||||
continue
|
||||
|
||||
try:
|
||||
img = Image.open(img_path)
|
||||
img_width, img_height = img.size
|
||||
except Exception as e:
|
||||
errors.append(f"无法打开图片: {img_file}")
|
||||
continue
|
||||
|
||||
with open(label_path, 'r') as f:
|
||||
for line_num, line in enumerate(f, 1):
|
||||
line = line.strip()
|
||||
if not line:
|
||||
continue
|
||||
|
||||
parts = line.split()
|
||||
if len(parts) != 5:
|
||||
errors.append(f"{label_file}: 格式错误")
|
||||
continue
|
||||
|
||||
try:
|
||||
class_id = int(parts[0])
|
||||
center_x = float(parts[1])
|
||||
center_y = float(parts[2])
|
||||
width = float(parts[3])
|
||||
height = float(parts[4])
|
||||
|
||||
if not (0 <= center_x <= 1 and 0 <= center_y <= 1):
|
||||
errors.append(f"{label_file}: 坐标超出范围")
|
||||
except ValueError:
|
||||
errors.append(f"{label_file}: 数字解析错误")
|
||||
|
||||
if errors:
|
||||
print("发现错误:")
|
||||
for error in errors[:10]:
|
||||
print(f" - {error}")
|
||||
else:
|
||||
print("验证通过!")
|
||||
|
||||
if __name__ == '__main__':
|
||||
validate_yolo_dataset('./dataset')
|
||||
```
|
||||
|
||||
运行验证:
|
||||
|
||||
```bash
|
||||
python validate_dataset.py
|
||||
```
|
||||
|
||||
### 4.2 创建数据集配置文件
|
||||
|
||||
创建 `dataset.yaml` 文件:
|
||||
|
||||
```yaml
|
||||
path: ./dataset
|
||||
train: images/train
|
||||
val: images/val
|
||||
test: images/test
|
||||
|
||||
nc: 3
|
||||
|
||||
names:
|
||||
0: car
|
||||
1: person
|
||||
2: bicycle
|
||||
```
|
||||
|
||||
**注意:** 根据你的实际类别修改 `nc` 和 `names`。
|
||||
|
||||
***
|
||||
|
||||
## 5. 数据集划分
|
||||
|
||||
### 5.1 创建划分脚本
|
||||
|
||||
创建 `split_dataset.py`:
|
||||
|
||||
```python
|
||||
import os
|
||||
import shutil
|
||||
import random
|
||||
|
||||
def split_dataset(source_dir, train_ratio=0.7, val_ratio=0.15, test_ratio=0.15, seed=42):
|
||||
random.seed(seed)
|
||||
|
||||
images_dir = os.path.join(source_dir, 'images')
|
||||
labels_dir = os.path.join(source_dir, 'labels')
|
||||
|
||||
images = [f for f in os.listdir(images_dir) if f.endswith(('.jpg', '.png'))]
|
||||
random.shuffle(images)
|
||||
|
||||
total = len(images)
|
||||
train_end = int(total * train_ratio)
|
||||
val_end = train_end + int(total * val_ratio)
|
||||
|
||||
train_images = images[:train_end]
|
||||
val_images = images[train_end:val_end]
|
||||
test_images = images[val_end:]
|
||||
|
||||
print(f"总图片数: {total}")
|
||||
print(f"训练集: {len(train_images)}")
|
||||
print(f"验证集: {len(val_images)}")
|
||||
print(f"测试集: {len(test_images)}")
|
||||
|
||||
for split, img_list in [('train', train_images), ('val', val_images), ('test', test_images)]:
|
||||
split_images_dir = os.path.join(source_dir, 'images', split)
|
||||
split_labels_dir = os.path.join(source_dir, 'labels', split)
|
||||
|
||||
os.makedirs(split_images_dir, exist_ok=True)
|
||||
os.makedirs(split_labels_dir, exist_ok=True)
|
||||
|
||||
for img in img_list:
|
||||
shutil.copy(os.path.join(images_dir, img), os.path.join(split_images_dir, img))
|
||||
|
||||
label_name = os.path.splitext(img)[0] + '.txt'
|
||||
src_label = os.path.join(labels_dir, label_name)
|
||||
dst_label = os.path.join(split_labels_dir, label_name)
|
||||
|
||||
if os.path.exists(src_label):
|
||||
shutil.copy(src_label, dst_label)
|
||||
|
||||
print("数据集划分完成!")
|
||||
|
||||
if __name__ == '__main__':
|
||||
split_dataset('./dataset')
|
||||
```
|
||||
|
||||
### 5.2 执行划分
|
||||
|
||||
```bash
|
||||
python split_dataset.py
|
||||
```
|
||||
|
||||
### 5.3 验证划分结果
|
||||
|
||||
```bash
|
||||
ls dataset/images/train
|
||||
ls dataset/images/val
|
||||
ls dataset/images/test
|
||||
```
|
||||
|
||||
***
|
||||
|
||||
## 6. 模型训练
|
||||
|
||||
### 6.1 创建训练脚本
|
||||
|
||||
创建 `train.py`:
|
||||
|
||||
```python
|
||||
from ultralytics import YOLO
|
||||
import torch
|
||||
|
||||
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
||||
print(f"使用设备: {device}")
|
||||
|
||||
model = YOLO('yolov8n.pt')
|
||||
|
||||
results = model.train(
|
||||
data='dataset.yaml',
|
||||
epochs=100,
|
||||
imgsz=640,
|
||||
batch=16,
|
||||
device=device,
|
||||
lr0=0.01,
|
||||
patience=50,
|
||||
save=True,
|
||||
plots=True,
|
||||
project='runs/detect',
|
||||
name='my_model',
|
||||
exist_ok=True,
|
||||
)
|
||||
|
||||
print("训练完成!")
|
||||
print(f"最佳模型保存在: runs/detect/my_model/weights/best.pt")
|
||||
```
|
||||
|
||||
### 6.2 开始训练
|
||||
|
||||
```bash
|
||||
python train.py
|
||||
```
|
||||
|
||||
### 6.3 训练参数说明
|
||||
|
||||
| 参数 | 说明 | 建议值 |
|
||||
| -------- | ------ | ----------- |
|
||||
| epochs | 训练轮数 | 100-300 |
|
||||
| imgsz | 输入图片尺寸 | 640 |
|
||||
| batch | 批次大小 | 根据 GPU 内存调整 |
|
||||
| lr0 | 初始学习率 | 0.01 |
|
||||
| patience | 早停耐心值 | 50 |
|
||||
|
||||
### 6.4 模型选择建议
|
||||
|
||||
| 模型 | 参数量 | 速度 | 精度 | 适用场景 |
|
||||
| ------- | ----- | -- | -- | ---- |
|
||||
| yolov8n | 3.2M | 最快 | 较低 | 实时检测 |
|
||||
| yolov8s | 11.2M | 快 | 中等 | 平衡 |
|
||||
| yolov8m | 25.9M | 中等 | 较高 | 生产环境 |
|
||||
| yolov8l | 43.7M | 较慢 | 高 | 高精度 |
|
||||
|
||||
***
|
||||
|
||||
## 7. 模型评估与测试
|
||||
|
||||
### 7.1 评估模型
|
||||
|
||||
创建 `evaluate.py`:
|
||||
|
||||
```python
|
||||
from ultralytics import YOLO
|
||||
|
||||
model = YOLO('runs/detect/my_model/weights/best.pt')
|
||||
|
||||
metrics = model.val(data='dataset.yaml', split='val')
|
||||
|
||||
print("=" * 50)
|
||||
print("模型评估结果")
|
||||
print("=" * 50)
|
||||
print(f"mAP50: {metrics.box.map50:.4f}")
|
||||
print(f"mAP50-95: {metrics.box.map:.4f}")
|
||||
print(f"Precision: {metrics.box.mp:.4f}")
|
||||
print(f"Recall: {metrics.box.mr:.4f}")
|
||||
print("=" * 50)
|
||||
```
|
||||
|
||||
运行评估:
|
||||
|
||||
```bash
|
||||
python evaluate.py
|
||||
```
|
||||
|
||||
### 7.2 测试单张图片
|
||||
|
||||
创建 `predict.py`:
|
||||
|
||||
```python
|
||||
from ultralytics import YOLO
|
||||
|
||||
model = YOLO('runs/detect/my_model/weights/best.pt')
|
||||
|
||||
results = model('dataset/images/test/test_image.jpg', save=True, conf=0.25)
|
||||
|
||||
for result in results:
|
||||
boxes = result.boxes
|
||||
for i in range(len(boxes)):
|
||||
class_name = model.names[int(boxes.cls[i])]
|
||||
conf = boxes.conf[i]
|
||||
print(f"检测到: {class_name}, 置信度: {conf:.2f}")
|
||||
```
|
||||
|
||||
运行测试:
|
||||
|
||||
```bash
|
||||
python predict.py
|
||||
```
|
||||
|
||||
### 7.3 批量测试
|
||||
|
||||
```python
|
||||
from ultralytics import YOLO
|
||||
|
||||
model = YOLO('runs/detect/my_model/weights/best.pt')
|
||||
|
||||
results = model('dataset/images/test', save=True, conf=0.25)
|
||||
```
|
||||
|
||||
### 7.4 性能基准
|
||||
|
||||
| 应用场景 | mAP50 目标 | mAP50-95 目标 |
|
||||
| ----- | -------- | ----------- |
|
||||
| 快速原型 | > 0.5 | > 0.3 |
|
||||
| 生产环境 | > 0.7 | > 0.5 |
|
||||
| 高精度应用 | > 0.9 | > 0.7 |
|
||||
|
||||
***
|
||||
|
||||
## 常见问题排查
|
||||
|
||||
### 问题 1:Loss 不下降
|
||||
|
||||
**解决方案:**
|
||||
|
||||
- 调整学习率(尝试 0.001-0.01)
|
||||
- 检查数据质量
|
||||
- 尝试更大的模型
|
||||
|
||||
### 问题 2:过拟合
|
||||
|
||||
**解决方案:**
|
||||
|
||||
- 增加数据量
|
||||
- 使用更小的模型
|
||||
- 启用更多数据增强
|
||||
|
||||
### 问题 3:训练很慢
|
||||
|
||||
**解决方案:**
|
||||
|
||||
- 使用 GPU 训练
|
||||
- 增大批次大小
|
||||
- 减小图片尺寸
|
||||
|
||||
***
|
||||
|
||||
## 目录结构总结
|
||||
|
||||
完成后的目录结构:
|
||||
|
||||
```
|
||||
project/
|
||||
├── venv/ # 虚拟环境目录(不应提交到版本控制)
|
||||
├── dataset/
|
||||
│ ├── images/
|
||||
│ │ ├── train/
|
||||
│ │ ├── val/
|
||||
│ │ └── test/
|
||||
│ └── labels/
|
||||
│ ├── train/
|
||||
│ ├── val/
|
||||
│ └── test/
|
||||
├── dataset.yaml
|
||||
├── train.py
|
||||
├── evaluate.py
|
||||
├── predict.py
|
||||
├── split_dataset.py
|
||||
└── validate_dataset.py
|
||||
```
|
||||
|
||||
***
|
||||
|
||||
## 快速开始命令汇总
|
||||
|
||||
```bash
|
||||
# 创建并激活虚拟环境
|
||||
python -m venv venv
|
||||
venv\Scripts\activate
|
||||
|
||||
# 安装依赖
|
||||
pip install ultralytics
|
||||
|
||||
# 后续步骤
|
||||
python validate_dataset.py
|
||||
python split_dataset.py
|
||||
python train.py
|
||||
python evaluate.py
|
||||
python predict.py
|
||||
```
|
||||
|
||||
***
|
||||
|
||||
## 一键运行环境准备
|
||||
|
||||
项目已提供 `run_demo.py` 脚本,可一键完成环境搭建:
|
||||
|
||||
```bash
|
||||
python run_demo.py
|
||||
```
|
||||
|
||||
该脚本会自动完成:
|
||||
|
||||
1. 检查 Python 版本
|
||||
2. 创建虚拟环境
|
||||
3. 安装 ultralytics 依赖
|
||||
4. 验证安装
|
||||
5. 检查 GPU 支持
|
||||
|
||||
## 在 VS Code 中直接运行代码
|
||||
|
||||
### 方法一:使用 Code Runner 插件
|
||||
|
||||
1. 打开 VS Code,搜索并安装 `Code Runner` 插件
|
||||
2. 将光标放在代码块内,点击右上角的 ▶️ 按钮运行
|
||||
|
||||
### 方法二:右键运行
|
||||
|
||||
在代码块内右键,选择 "Run Code" 或使用快捷键 `Ctrl+Alt+N`
|
||||
|
||||
### 方法三:终端运行
|
||||
|
||||
```bash
|
||||
# 进入项目目录
|
||||
cd .
|
||||
|
||||
# 激活虚拟环境
|
||||
venv\Scripts\activate
|
||||
|
||||
# 运行脚本
|
||||
python run_demo.py
|
||||
python validate_dataset.py
|
||||
python split_dataset.py
|
||||
python train.py
|
||||
```
|
||||
|
||||
***
|
||||
|
||||
祝训练顺利!
|
||||
Reference in New Issue
Block a user