high-quality-midjouney-srefs

# Midjourney Image Scraper & Dataset Creator

A complete toolkit for scraping Midjourney images, generating captions, and creating HuggingFace datasets with optional automatic upload to HuggingFace Hub.

## 🌟 Features

- **🔍 Web Scraping**: Download images from midjourneysref.com with comprehensive error handling
- **🤖 AI Captioning**: Automatic image captioning using Moondream API with **auto-resume capability**
- **✂️ Smart Cropping**: AI-powered image cropping using OpenAI to optimize aspect ratios
- **🔄 Interrupt-Safe**: Stop and restart any process - it'll pick up where you left off
- **📦 HuggingFace Datasets**: Create professional HF-compatible datasets with metadata
- **☁️ Hub Upload**: Direct upload to HuggingFace Hub with one command
- **📊 Detailed Logging**: Comprehensive feedback and statistics for all operations

## 🚀 Quick Start

### 1. Installation

```bash
# Clone or download this repository
git clone <your-repo-url>
cd style_scraper

# Install all dependencies
pip install -r requirements.txt
```

### 2. Environment Setup

Create a `.env` file in the project root:

```bash
# Required for image captioning
MOONDREAM_API_KEY=your_moondream_api_key_here

# Required for AI-powered cropping
OPENAI_API_KEY=your_openai_api_key_here

# Required for HuggingFace Hub upload
HF_TOKEN=your_huggingface_token_here
```

**Get your tokens:**
- **Moondream API**: https://moondream.ai/api
- **OpenAI API**: https://platform.openai.com/api-keys
- **HuggingFace Hub**: https://huggingface.co/settings/tokens (select "Write" permissions)

### 3. Basic Usage

```bash
# 1. Scrape images
python scraper.py

# 2. Generate captions (optional but recommended)
python caption_images.py

# 3. Crop images with AI analysis (optional)
python crop_dataset.py

# 4. Create HuggingFace dataset
python create_hf_dataset.py --prompts prompts.csv

# 5. Or create and upload in one step
python create_hf_dataset.py --prompts prompts.csv --upload --repo-id your-username/my-dataset
```

## 📖 Detailed Usage

### Step 1: Image Scraping (`scraper.py`)

Downloads images from midjourneysref.com with comprehensive error handling.

```bash
python scraper.py
```

**Features:**
- Downloads images from pages 6-20 by default (configurable in script)
- Pre-flight connectivity testing
- Duplicate detection and skipping
- Detailed error categorization and statistics
- Comprehensive logging to `scraper.log`

**Output:** Images saved to `midjourney_images/` folder

### Step 2: Image Captioning (`caption_images.py`)

Generates captions for uncaptioned images using the Moondream API. **Automatically resumes** from where you left off!

```bash
# Basic usage - automatically resumes from existing captions
python caption_images.py

# First run captions 50 images, then gets interrupted...
# Second run automatically continues with remaining images
python caption_images.py # No extra flags needed!

# Force recaption all images (ignoring existing captions)
python caption_images.py --recaption

# Custom options
python caption_images.py --images my_images/ --output captions.csv --delay 0.5

# Caption length control
python caption_images.py --prompt-length short # Concise captions
python caption_images.py --prompt-length normal # Detailed captions
python caption_images.py --prompt-length mix # Random mix (default)
```

**Options:**
- `--images, -i`: Input folder (default: `midjourney_images`)
- `--output, -o`: Output CSV file (default: `prompts.csv`)
- `--existing, -e`: Custom existing prompts file (optional, by default uses output file)
- `--recaption, -r`: Force recaption all images, ignoring existing captions
- `--prompt-length, -l`: Caption length - `short`, `normal`, or `mix` (default: `mix`)
- `--delay, -d`: API rate limiting delay (default: 1.0 seconds)
- `--verbose, -v`: Enable debug logging

**Features:**
- **🔄 Auto-resume**: Automatically skips already-captioned images by default
- **🚀 Interrupt-safe**: Can safely stop and restart the process anytime
- Rate limiting to respect API limits
- API connection testing before processing
- Comprehensive error handling and statistics
- Compatible CSV output for dataset creation

**Output:** CSV file with `filename,prompt` columns

### Step 3: AI-Powered Image Cropping (`crop_dataset.py`)

Analyzes images using OpenAI's vision API to determine optimal crop ratios and creates cropped versions.

```bash
# Basic usage - processes all images in prompts.csv
python crop_dataset.py

# Test with a single image first
python test_crop.py
```

**Features:**
- **🤖 AI Analysis**: Uses OpenAI GPT-4 Vision to analyze each image for optimal cropping
- **📐 Smart Ratios**: Supports 16:9, 9:16, 4:3, 3:4, 1:1, or no cropping
- **🎯 Content-Aware**: Considers image content, composition, and aesthetics
- **📊 Metadata Tracking**: Saves crop ratios and dimensions to updated CSV
- **🔒 Fallback Safe**: Defaults to 1:1 (square) for invalid responses or errors

**Crop Ratios:**
- `16:9` - Wide scenes, landscapes, group shots
- `9:16` - Tall subjects, full-body portraits, buildings
- `4:3` - Balanced framing, single subjects, general scenes
- `3:4` - Portraits, vertical objects, close-ups
- `1:1` - Square format, where entire image is important
- `no` - Keep original aspect ratio (treated as 1:1)

**Output:**
- `cropped_images/` - Folder with all cropped images
- `prompts_with_crops.csv` - Updated CSV with crop metadata
- `crop_dataset.log` - Detailed processing log

**CSV Columns Added:**
- `crop_ratio` - AI-recommended aspect ratio
- `cropped_width` - Width of cropped image in pixels
- `cropped_height` - Height of cropped image in pixels
- `cropped_filename` - Filename of the cropped version

### Step 4: Dataset Creation (`create_hf_dataset.py`)

Creates a professional HuggingFace-compatible dataset with metadata extraction.

```bash
# Local dataset only
python create_hf_dataset.py --prompts prompts.csv

# Using cropped images and metadata
python create_hf_dataset.py --prompts prompts_with_crops.csv --input cropped_images

# With HuggingFace Hub upload
python create_hf_dataset.py --prompts prompts.csv --upload --repo-id username/dataset-name
```

**Options:**
- `--input, -i`: Input images folder (default: `midjourney_images`)
- `--output, -o`: Output dataset folder (default: `midjourney_hf_dataset`)
- `--name, -n`: Dataset name (default: `midjourney-images`)
- `--prompts, -p`: Path to prompts CSV file
- `--upload, -u`: Upload to HuggingFace Hub
- `--repo-id, -r`: HuggingFace repository ID (e.g., `username/dataset-name`)
- `--verbose, -v`: Enable debug logging

**Features:**
- Extracts comprehensive image metadata (resolution, file size, orientation, etc.)
- Style reference (`sref`) extraction from filenames
- HuggingFace-compatible structure and configuration
- Automatic README generation with usage examples
- Dataset loading scripts for seamless integration
- Optional direct upload to HuggingFace Hub

## 📊 Output Structure

### Dataset Directory Structure
```
midjourney_hf_dataset/
├── images/ # All image files
├── metadata/
│ ├── metadata.csv # Main HuggingFace metadata
│ ├── metadata_detailed.json # Detailed metadata
│ └── dataset_summary.json # Dataset statistics
├── dataset_config/ # HuggingFace configuration
│ ├── dataset_infos.json
│ └── midjourney-images.py # Dataset loading script
└── README.md # Generated documentation
```

### Metadata Fields

Each image includes comprehensive metadata:

- `filename`: Original image filename
- `file_path`: Relative path within dataset
- `sref`: Style reference ID (extracted from filename)
- `prompt`: AI-generated or provided caption
- `width`: Image width in pixels
- `height`: Image height in pixels
- `file_size_mb`: File size in megabytes
- `size_category`: Resolution category (high/medium/low)
- `orientation`: Image orientation (landscape/portrait/square)

## 🔧 Configuration

### Scraper Configuration

Edit variables in `scraper.py`:

```python
# Folder where you want to save the images
DOWNLOAD_FOLDER = "midjourney_images"

# Page range to scrape
start_page = 6
end_page = 20
```

### Caption Configuration

The captioning script supports rate limiting, automatic resuming, and caption length control:

```bash
# Caption length options
python caption_images.py --prompt-length short # Short, concise captions
python caption_images.py --prompt-length normal # Detailed descriptions
python caption_images.py --prompt-length mix # Random mix (default - adds variety)

# Rate limiting for API stability
python caption_images.py --delay 1.0

# Auto-resume from previous run (default behavior)
python caption_images.py

# Force recaption everything from scratch
python caption_images.py --recaption

# Use a different existing file for comparison
python caption_images.py --existing other_captions.csv
```

### Dataset Configuration

Customize dataset creation:

```bash
# Custom dataset name and structure
python create_hf_dataset.py \
--input my_images/ \
--output my_custom_dataset/ \
--name "my-custom-midjourney-dataset" \
--prompts captions.csv
```

## 🤗 Using Your Dataset

### Loading from HuggingFace Hub

```python
from datasets import load_dataset

# Load your uploaded dataset
dataset = load_dataset("username/your-dataset-name")

# Access data
for example in dataset["train"]:
image = example["image"] # PIL Image
prompt = example["prompt"] # Caption text
sref = example["sref"] # Style reference
width = example["width"] # Image width
# ... other metadata
```

### Loading Locally

```python
import pandas as pd
from PIL import Image
import os

# Load metadata
metadata = pd.read_csv("midjourney_hf_dataset/metadata/metadata.csv")

# Load specific image
def load_image(filename):
return Image.open(f"midjourney_hf_dataset/images/{filename}")

# Filter by criteria
high_res = metadata[metadata['size_category'] == 'high_resolution']
has_prompts = metadata[metadata['prompt'] != ""]
same_style = metadata[metadata['sref'] == '4160600070']
```

## 📈 Advanced Usage

### Batch Processing

```bash
# Process multiple scraping sessions - auto-resumes captioning!
for i in {1..5}; do
python scraper.py
python caption_images.py # Automatically skips existing captions
done

# Create final dataset
python create_hf_dataset.py --prompts prompts.csv --upload --repo-id username/large-midjourney-dataset

# Alternative: Force fresh captions for each batch
for i in {1..5}; do
python scraper.py
python caption_images.py --recaption
done
```

### Custom Prompts

You can provide your own prompts instead of using AI captioning:

```csv
filename,prompt
4160600070-1-d9409ee5.png,"A majestic dragon soaring over snow-capped mountains"
4160600070-2-a8b7c9d2.png,"Cyberpunk cityscape with neon reflections in rain"
```

### Style-Based Datasets

```bash
# Filter by style reference before creating dataset
python -c "
import pandas as pd
df = pd.read_csv('prompts.csv')
style_df = df[df['filename'].str.startswith('4160600070')]
style_df.to_csv('style_specific_prompts.csv', index=False)
"

python create_hf_dataset.py --prompts style_specific_prompts.csv --name "style-4160600070"
```

## 🐛 Troubleshooting

### Common Issues

**1. Missing API Keys**
```
Error: MOONDREAM_API_KEY not found
```
- Ensure `.env` file exists with valid API key
- Check API key has sufficient credits

**2. HuggingFace Upload Fails**
```
Error: HF_TOKEN not found
```
- Create token at https://huggingface.co/settings/tokens
- Ensure "Write" permissions are selected
- Check repository name is available

**3. No Images Found**
```
Warning: No images found with primary selector
```
- Website structure may have changed
- Check internet connection
- Verify target pages exist

**4. Caption Generation Fails**
```
Failed to caption image: API error
```
- Check Moondream API status
- Verify API key and credits
- Reduce rate limiting with `--delay`
- The script auto-resumes, so you can safely restart after fixing the issue

**5. Want to Recaption Existing Images**
```
Images already have captions but I want to regenerate them
```
- Use `--recaption` flag to ignore existing captions
- Or delete the existing CSV file to start fresh

### Log Files

Check these log files for detailed debugging:
- `scraper.log`: Web scraping logs
- `caption_images.log`: Captioning process logs
- `hf_dataset_creation.log`: Dataset creation logs

## 📄 License

This project is for educational and research purposes. Please respect the terms of service of the source website and API providers.

## 🤝 Contributing

Feel free to submit issues and enhancement requests!

## 🙏 Acknowledgments

- Images sourced from [midjourneysref.com](https://midjourneysref.com)
- Captioning powered by [Moondream API](https://moondream.ai)
- Dataset hosting by [🤗 Hugging Face](https://huggingface.co)

# Midjourney 图像抓取器与数据集创建工具（Midjourney Image Scraper & Dataset Creator）

一款用于抓取Midjourney图像、生成图像标题以及创建HuggingFace数据集的完整工具包，支持可选的自动上传至HuggingFace Hub。

## 🌟 核心特性

- **🔍 网页抓取**：从midjourneysref.com下载图像，附带全面的错误处理机制
- **🤖 AI 图像标题生成**：使用Moondream API自动生成图像标题，支持**自动续跑功能**
- **✂️ 智能裁剪**：依托OpenAI的AI能力进行图像裁剪，优化画幅比例
- **🔄 断点续跑**：可随时停止或重启进程，程序会从上次中断处自动恢复执行
- **📦 HuggingFace 数据集**：创建符合专业标准的HF兼容数据集并附带元数据
- **☁️ Hub 上传**：单命令直接上传至HuggingFace Hub
- **📊 详细日志**：为所有操作提供全面的反馈与统计信息

## 🚀 快速上手

### 1. 安装部署

bash
# 克隆或下载本仓库
git clone <your-repo-url>
cd style_scraper

# 安装所有依赖项
pip install -r requirements.txt

### 2. 环境配置

在项目根目录创建 `.env` 文件：

bash
# 图像标题生成所需
MOONDREAM_API_KEY=your_moondream_api_key_here

# AI 图像裁剪所需
OPENAI_API_KEY=your_openai_api_key_here

# HuggingFace Hub 上传所需
HF_TOKEN=your_huggingface_token_here

**获取对应令牌：**
- **Moondream API**：https://moondream.ai/api
- **OpenAI API**：https://platform.openai.com/api-keys
- **HuggingFace Hub**：https://huggingface.co/settings/tokens（需选择“写入”权限）

### 3. 基础用法

bash
# 1. 抓取图像
python scraper.py

# 2. 生成图像标题（可选但推荐）
python caption_images.py

# 3. 使用AI分析裁剪图像（可选）
python crop_dataset.py

# 4. 创建HuggingFace数据集
python create_hf_dataset.py --prompts prompts.csv

# 5. 或一步完成创建与上传
python create_hf_dataset.py --prompts prompts.csv --upload --repo-id your-username/my-dataset

## 📖 详细使用指南

### 步骤1：图像抓取（`scraper.py`）

从midjourneysref.com下载图像，附带全面的错误处理机制。

bash
python scraper.py

**功能特性：**
- 默认抓取第6至20页的内容（可在脚本中自定义配置）
- 预检连接测试
- 重复图像检测与跳过
- 详细的错误分类与统计信息
- 日志文件写入至 `scraper.log`

**输出：** 图像保存至 `midjourney_images/` 文件夹

### 步骤2：图像标题生成（`caption_images.py`）

使用Moondream API为未标注标题的图像自动生成标题，**支持自动续跑**，从上次中断处恢复。

bash
# 基础用法 - 自动从已有标题处续跑
python caption_images.py

# 首次运行生成50张图像的标题后中断...
# 第二次运行将自动继续处理剩余图像
python caption_images.py # 无需额外参数！

# 强制重新为所有图像生成标题（忽略已有标题）
python caption_images.py --recaption

# 自定义选项
python caption_images.py --images my_images/ --output captions.csv --delay 0.5

# 标题长度控制
python caption_images.py --prompt-length short # 简短精炼的标题
python caption_images.py --prompt-length normal # 详细描述性标题
python caption_images.py --prompt-length mix # 随机混合模式（默认）

**可选参数：**
- `--images, -i`：输入图像文件夹（默认：`midjourney_images`）
- `--output, -o`：输出CSV文件路径（默认：`prompts.csv`）
- `--existing, -e`：自定义已有标题文件（可选，默认使用输出文件）
- `--recaption, -r`：强制重新生成所有图像标题，忽略已有内容
- `--prompt-length, -l`：标题长度模式 - `short`、`normal` 或 `mix`（默认：`mix`）
- `--delay, -d`：API请求速率限制延迟（默认：1.0秒）
- `--verbose, -v`：启用调试日志

**功能特性：**
- **🔄 自动续跑**：默认自动跳过已生成标题的图像
- **🚀 断点安全**：可随时安全停止或重启进程
- 速率限制以遵守API调用限制
- 处理前的API连接测试
- 全面的错误处理与统计信息
- 兼容数据集创建所需的CSV输出格式

**输出：** 包含`filename,prompt`两列的CSV文件

### 步骤3：AI驱动的图像裁剪（`crop_dataset.py`）

使用OpenAI视觉API分析图像，确定最优裁剪比例并生成裁剪后的图像版本。

bash
# 基础用法 - 处理prompts.csv中的所有图像
python crop_dataset.py

# 先单张图像测试裁剪效果
python test_crop.py

**功能特性：**
- **🤖 AI 分析**：使用OpenAI GPT-4 Vision分析每张图像，确定最优裁剪方案
- **📐 智能比例**：支持16:9、9:16、4:3、3:4、1:1或不裁剪模式
- **🎯 内容感知**：考虑图像内容、构图与美学效果
- **📊 元数据跟踪**：将裁剪比例与尺寸信息保存至更新后的CSV文件
- **🔒 安全 fallback**：若API返回无效结果或发生错误，默认使用1:1（方形）裁剪

**支持的裁剪比例：**
- `16:9` - 宽幅场景、风景、群组合影
- `9:16` - 竖幅主体、全身人像、建筑拍摄
- `4:3` - 均衡构图、单一主体、通用场景
- `3:4` - 人像、垂直物体、特写拍摄
- `1:1` - 方形画幅，需保留完整图像内容
- `no` - 保留原始画幅比例（视为1:1处理）

**输出：**
- `cropped_images/` - 存放所有裁剪后图像的文件夹
- `prompts_with_crops.csv` - 包含裁剪元数据的更新版CSV文件
- `crop_dataset.log` - 详细的处理日志

**新增CSV列：**
- `crop_ratio` - AI推荐的画幅比例
- `cropped_width` - 裁剪后图像的像素宽度
- `cropped_height` - 裁剪后图像的像素高度
- `cropped_filename` - 裁剪后图像的文件名

### 步骤4：数据集创建（`create_hf_dataset.py`）

创建符合专业标准的HuggingFace兼容数据集并提取元数据。

bash
# 仅生成本地数据集
python create_hf_dataset.py --prompts prompts.csv

# 使用裁剪后的图像与元数据
python create_hf_dataset.py --prompts prompts_with_crops.csv --input cropped_images

# 直接上传至HuggingFace Hub
python create_hf_dataset.py --prompts prompts.csv --upload --repo-id username/dataset-name

**可选参数：**
- `--input, -i`：输入图像文件夹（默认：`midjourney_images`）
- `--output, -o`：输出数据集文件夹（默认：`midjourney_hf_dataset`）
- `--name, -n`：数据集名称（默认：`midjourney-images`）
- `--prompts, -p`：标题CSV文件路径
- `--upload, -u`：上传至HuggingFace Hub
- `--repo-id, -r`：HuggingFace仓库ID（例如：`username/dataset-name`）
- `--verbose, -v`：启用调试日志

**功能特性：**
- 提取全面的图像元数据（分辨率、文件大小、方向等）
- 从文件名提取风格参考ID（sref）
- 符合HuggingFace规范的目录结构与配置
- 自动生成包含使用示例的README文档
- 数据集加载脚本，实现无缝集成
- 可选直接上传至HuggingFace Hub

## 📊 输出目录结构

### 数据集目录结构

midjourney_hf_dataset/
├── images/ # 所有图像文件
├── metadata/
│ ├── metadata.csv # HuggingFace主元数据文件
│ ├── metadata_detailed.json # 详细元数据
│ └── dataset_summary.json # 数据集统计信息
├── dataset_config/ # HuggingFace配置文件
│ ├── dataset_infos.json
│ └── midjourney-images.py # 数据集加载脚本
└── README.md # 自动生成的文档

### 元数据字段

每张图像包含以下全面元数据：

- `filename`：原始图像文件名
- `file_path`：数据集内的相对路径
- `sref`：风格参考ID（从文件名提取）
- `prompt`：AI生成或提供的图像标题
- `width`：图像像素宽度
- `height`：图像像素高度
- `file_size_mb`：文件大小（单位：MB）
- `size_category`：分辨率类别（高/中/低）
- `orientation`：图像方向（横向/纵向/方形）

## 🔧 配置调整

### 抓取脚本配置

编辑`scraper.py`中的变量：

python
# 图像保存文件夹
DOWNLOAD_FOLDER = "midjourney_images"

# 抓取的页码范围
start_page = 6
end_page = 20

### 标题生成脚本配置

标题生成脚本支持速率限制、自动续跑与标题长度控制：

bash
# 标题长度选项
python caption_images.py --prompt-length short # 简短标题
python caption_images.py --prompt-length normal # 详细描述标题
python caption_images.py --prompt-length mix # 随机混合模式（默认，增加多样性）

# 调整API请求速率以保证稳定性
python caption_images.py --delay 1.0

# 从上次运行处自动续跑（默认行为）
python caption_images.py

# 强制从头开始重新生成所有标题
python caption_images.py --recaption

# 使用其他已有标题文件进行比对
python caption_images.py --existing other_captions.csv

### 数据集创建脚本配置

自定义数据集创建参数：

bash
# 自定义数据集名称与结构
python create_hf_dataset.py
--input my_images/
--output my_custom_dataset/
--name "my-custom-midjourney-dataset"
--prompts captions.csv

## 🤗 使用生成的数据集

### 从HuggingFace Hub加载

python
from datasets import load_dataset

# 加载您上传的数据集
dataset = load_dataset("username/your-dataset-name")

# 访问数据
for example in dataset["train"]:
image = example["image"] # PIL Image图像对象
prompt = example["prompt"] # 图像标题文本
sref = example["sref"] # 风格参考ID
width = example["width"] # 图像宽度
# ... 其他元数据字段

### 本地加载数据集

python
import pandas as pd
from PIL import Image
import os

# 加载元数据
metadata = pd.read_csv("midjourney_hf_dataset/metadata/metadata.csv")

# 加载指定图像
def load_image(filename):
return Image.open(f"midjourney_hf_dataset/images/{filename}")

# 按条件筛选数据
high_res = metadata[metadata['size_category'] == 'high_resolution']
has_prompts = metadata[metadata['prompt'] != ""]
same_style = metadata[metadata['sref'] == '4160600070']

## 📈 高级用法

### 批量处理

bash
# 多次抓取会话 - 标题生成自动续跑！
for i in {1..5}; do
python scraper.py
python caption_images.py # 自动跳过已生成标题的图像
done

# 创建最终数据集
python create_hf_dataset.py --prompts prompts.csv --upload --repo-id username/large-midjourney-dataset

# 替代方案：为每个批处理重新生成标题
for i in {1..5}; do
python scraper.py
python caption_images.py --recaption
done

### 自定义标题

您可以提供自定义标题而非使用AI生成：

csv
filename,prompt
4160600070-1-d9409ee5.png,"A majestic dragon soaring over snow-capped mountains"
4160600070-2-a8b7c9d2.png,"Cyberpunk cityscape with neon reflections in rain"

### 基于风格的专属数据集

bash
# 在创建数据集前按风格参考ID筛选数据
python -c "
import pandas as pd
df = pd.read_csv('prompts.csv')
style_df = df[df['filename'].str.startswith('4160600070')]
style_df.to_csv('style_specific_prompts.csv', index=False)
"

python create_hf_dataset.py --prompts style_specific_prompts.csv --name "style-4160600070"

## 🐛 故障排除

### 常见问题

**1. 缺少API密钥**

Error: MOONDREAM_API_KEY not found

- 确保`.env`文件存在且包含有效的API密钥
- 检查API密钥是否有足够的调用额度

**2. HuggingFace上传失败**

Error: HF_TOKEN not found

- 在https://huggingface.co/settings/tokens创建令牌
- 确保选择了“写入”权限
- 检查仓库名称是否可用

**3. 未找到任何图像**

Warning: No images found with primary selector

- 目标网站的结构可能已变更
- 检查网络连接
- 确认目标页面是否存在

**4. 标题生成失败**

Failed to caption image: API error

- 检查Moondream API的运行状态
- 确认API密钥与调用额度是否有效
- 使用`--delay`参数降低请求速率
- 脚本支持自动续跑，修复问题后可安全重启

**5. 需要重新生成已有图像的标题**

Images already have captions but I want to regenerate them

- 使用`--recaption`参数忽略已有标题重新生成
- 或删除现有CSV文件从头开始

### 日志文件

查看以下日志文件进行详细调试：
- `scraper.log`：网页抓取日志
- `caption_images.log`：标题生成进程日志
- `hf_dataset_creation.log`：数据集创建日志

## 📄 许可证

本项目仅用于教育与研究用途，请遵守源网站与API服务提供商的服务条款。

## 🤝 贡献

欢迎提交问题报告与功能改进请求！

## 🙏 致谢

- 图像来源：[midjourneysref.com](https://midjourneysref.com)
- 标题生成服务：[Moondream API](https://moondream.ai)
- 数据集托管：[🤗 Hugging Face](https://huggingface.co)