https://github.com/pathwiselabs/pixel-pipeline
A Python application with Gradio UI for batch processing and captioning of images, allowing for easy integration with AI image training workflows.
https://github.com/pathwiselabs/pixel-pipeline
data-cleaning data-science flux generative-ai stable-diffusion stable-diffusion-webui
Last synced: 4 months ago
JSON representation
A Python application with Gradio UI for batch processing and captioning of images, allowing for easy integration with AI image training workflows.
- Host: GitHub
- URL: https://github.com/pathwiselabs/pixel-pipeline
- Owner: pathwiselabs
- Created: 2025-04-17T08:25:19.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-05-20T20:32:14.000Z (about 1 year ago)
- Last Synced: 2025-06-15T22:40:19.075Z (about 1 year ago)
- Topics: data-cleaning, data-science, flux, generative-ai, stable-diffusion, stable-diffusion-webui
- Language: Python
- Homepage:
- Size: 403 KB
- Stars: 7
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
## What is it?
Pixel Pipeline is a desktop application that helps you refine image datasets for AI training by cleaning, filtering, and enhancing your image collections. It offers a comprehensive set of tools for removing duplicates, filtering images by face count, reducing dataset size while maintaining diversity, and generating high-quality AI captions.
## Features
### Complete Image Dataset Refinement Workflow
- **Image Similarity Detection**: Remove duplicates and visually similar images
- Perceptual hash for exact duplicates
- VGG16-based similarity detection for visually similar images
- **Face Detection & Filtering**: Sort images by face count
- Multiple detection algorithms (MTCNN, MediaPipe, YuNet, etc.)
- Automatic sorting into no-face, single-face, and multiple-face categories
- **Image Set Refinement**: Intelligently reduce dataset size
- Face embedding and clustering
- Diversity or consistency-based selection
- Interactive visualization of clustering results
- **AI-Powered Image Captioning**: Generate high-quality captions
- Qwen2.5-VL vision-language models (3B & 7B variants)
- Custom prompting for tailored descriptions
- Flux-compatible identifier generation
- Caption export as ZIP files
## System Requirements
- **Python**: 3.11.9 or compatible
- **CUDA**: 12.8 (Blackwell compatible)
- **PyTorch**: Nightly builds of torch and torchvision (Blackwell compatible)
- **RAM**: 8GB minimum, 16GB+ recommended
- **GPU**: NVIDIA GPU with 6GB+ VRAM recommended for optimal performance
- **Storage**: 10GB+ free space for models and temporary files
## Installation
1. Clone this repository:
```bash
git clone https://github.com/yourusername/pixel-pipeline.git
cd pixel-pipeline
```
2. Create and activate a virtual environment:
```bash
python -m venv venv
# On Windows
venv\Scripts\activate
# On macOS/Linux
source venv/bin/activate
```
3. Install the required dependencies:
```bash
pip install -r requirements.txt
```
**Note for CUDA compatibility**: This project requires CUDA 12.8 compatible PyTorch. The requirements.txt file includes the correct version specifications.
## Usage
### Launching the Application
Run the application using the included batch file (recommended):
```bash
run.bat
```
Or launch directly with Python:
```bash
python app.py
```
The application will open in your default web browser at http://localhost:7865.
### Recommended Workflow
For best results, follow this sequence:
1. **Image Similarity Tab**: First remove duplicates and overly similar images
2. **Face Detection Tab**: Next, filter images based on face count (for character LoRA training)
3. **Image Set Refinement Tab**: Reduce dataset size while maintaining diversity or consistency
4. **Image Captioning Tab**: Finally, add AI-generated captions to your refined dataset
## FAQ
### Why am I getting CUDA/GPU errors?
Ensure you have NVIDIA drivers installed that support CUDA 12.8. You can also try using a smaller batch size or the 3B model variant for captioning if you have limited VRAM.
### Can I run this without a GPU?
Yes, but performance will be significantly slower, especially for face detection, clustering, and captioning tasks.
### How do I report a bug or request a feature?
Please open an issue on the GitHub repository with a detailed description of the bug or feature request.
### What image formats are supported?
Pixel Pipeline supports JPG, PNG, and JPEG image formats.
### How many images can I process at once?
This depends on your hardware. With sufficient RAM and VRAM, you can process hundreds or thousands of images, but for optimal performance, consider batching large datasets.
## Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
1. Fork the repository
2. Create your feature branch (`git checkout -b feature/amazing-feature`)
3. Commit your changes (`git commit -m 'Add some amazing feature'`)
4. Push to the branch (`git push origin feature/amazing-feature`)
5. Open a Pull Request
## License
MIT License
## Acknowledgments
- This application utilizes the [Qwen2.5-VL](https://github.com/QwenLM/Qwen-VL) model from Alibaba
- Built with [Gradio](https://www.gradio.app/) for the user interface
- [Hugging Face Transformers](https://huggingface.co/docs/transformers/index) for model integration
- Face detection and analysis powered by various open-source libraries including DeepFace and FaceNet