File size: 4,740 Bytes
23c2970 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 | ---
license: apache-2.0
---
# WithoutBG Snap Models
**Free, high-quality background removal models powered by AI**
[](https://opensource.org/licenses/Apache-2.0)
[](https://onnxruntime.ai/)
This repository contains the **Snap tier** models for the [withoutbg](https://github.com/withoutbg/withoutbg) library - a complete set of ONNX models for local, free background removal processing.
## 📋 Model Overview
The Snap tier implements a sophisticated 3-stage pipeline for background removal:
| Model | File | Purpose | Input | Output | License |
|-------|------|---------|-------|---------|---------|
| **Depth Estimation** | `depth_anything_v2_vits_slim.onnx` | Stage 1: Depth map generation | RGB (518×518) | Inverse depth map | Apache 2.0 |
| **Matting** | `snap_matting_0.1.0.onnx` | Stage 2: Initial background separation | RGBD (256×256) | Alpha channel (A1) | Apache 2.0 |
| **Refiner** | `snap_refiner_0.1.0.onnx` | Stage 3: High-resolution refinement | RGB+D+A (original size) | Refined alpha (A2) | Apache 2.0 |
## 🚀 Quick Start
### Using the withoutbg library (Recommended)
```bash
pip install withoutbg
```
```python
from withoutbg import remove_background
# Automatically downloads and uses these models
result = remove_background("image.jpg")
result.save("output.png")
```
## 🔄 Processing Pipeline
### Stage 1: Depth Estimation
- **Model**: Depth Anything V2 ViT-S (Apache 2.0 licensed)
- **Input**: RGB image (518×518 pixels, ImageNet normalized)
- **Output**: Inverse depth map (0-255 range)
- **Purpose**: Provides spatial understanding for better background separation
### Stage 2: Matting
- **Input**: RGBD (RGB + depth concatenated as 4-channel input, 256×256)
- **Output**: Initial alpha channel (A1)
- **Purpose**: Performs initial foreground/background segmentation
### Stage 3: Refining
- **Input**: RGB + depth + A1 (5-channel input at original resolution)
- **Output**: Refined alpha channel (A2) with high detail
- **Purpose**: Enhances edge quality and removes artifacts
## 🔧 Technical Specifications
### Model Details
- **Framework**: ONNX (compatible with ONNX Runtime)
- **Providers**: CPU, CUDA (automatically detected)
- **Precision**: FP32
- **Total Size**: ~140 MB (all three models)
### Input Requirements
- **Format**: RGB images (any resolution)
- **Preprocessing**: Automatic resizing and normalization
- **Output**: RGBA images with transparent background
## 🏗️ Integration Examples
### Batch Processing
```python
from withoutbg import remove_background_batch
results = remove_background_batch([
"image1.jpg",
"image2.jpg",
"image3.jpg"
], output_dir="results/")
```
### Custom Model Paths
```python
from withoutbg.models import SnapModel
model = SnapModel(
depth_model_path="custom/depth_model.onnx",
matting_model_path="custom/matting_model.onnx",
refiner_model_path="custom/refiner_model.onnx"
)
result = model.remove_background("image.jpg")
```
## 📄 Licensing
### Open Source Components
- **`depth_anything_v2_vits_slim.onnx`**: Apache 2.0 License
- Based on [Depth-Anything V2](https://github.com/DepthAnything/Depth-Anything-V2)
### Snap Tier Components
- **`snap_matting_0.1.0.onnx`**: Apache 2.0 License
- **`snap_refiner_0.1.0.onnx`**: Apache 2.0 License
- Free for commercial and non-commercial use
- Open source models for the Snap tier
## 🔗 Related Links
- **Main Library**: [withoutbg/withoutbg](https://github.com/withoutbg/withoutbg)
- **Documentation**: [withoutbg.com/docs](https://withoutbg.com/documentation)
- **Demo**: [Hugging Face Space](https://huggingface.co/spaces/withoutbg/demo)
- **Commercial Licensing**: [withoutbg.com/focus](https://withoutbg.com/focus)
## 🎯 Use Cases
- **Development & Prototyping**: Free local processing
- **E-commerce**: Product photo background removal
- **Social Media**: Profile picture editing
- **Content Creation**: Video thumbnails and graphics
## 🤝 Contributing
We welcome improvements to the open source components:
1. **Depth Anything V2 optimizations**: Submit PRs to improve inference speed
2. **Preprocessing enhancements**: Better image handling and normalization
3. **Documentation**: Examples, tutorials, and integration guides
4. **Bug reports**: Issues with model loading or inference
## 📧 Support
- **Technical Issues**: [GitHub Issues](https://github.com/withoutbg/withoutbg/issues)
- **Community**: [GitHub Discussions](https://github.com/withoutbg/withoutbg/discussions)
---
**Try the models**: Install `pip install withoutbg` and start removing backgrounds instantly! |