README.md

---
license: mit
tags:
- depth-estimation
- onnx
- computer-vision
- visiondepth3d
- mit-license
---

# Distill-Any-Depth-Small (ONNX) – For VisionDepth3D

> **Model Origin:** This model is based on [Distill-Any-Depth by Westlake-AGI-Lab](https://github.com/Westlake-AGI-Lab/Distill-Any-Depth), originally developed by Westlake-AGI-Lab.  
> I did not train this model — I have converted it to ONNX format for fast, GPU-accelerated inference within tools such as VisionDepth3D.

## 🧠 About This Model

This is a direct conversion of the **Distill-Any-Depth** PyTorch model to **ONNX**, intended for lightweight, real-time depth estimation from single RGB images.

### ✔️ Key Features:
- ONNX format (exported from PyTorch)
- Compatible with ONNX Runtime and TensorRT
- Excellent for 2D to 3D depth workflows
- Works seamlessly with **VisionDepth3D**

## 📌 Intended Use

- Real-time or batch depth map generation
- 2D to 3D conversion pipelines (e.g., SBS 3D video)
- Works on Windows, Linux (CUDA-supported)

## 📜 License and Attribution

### Citation

```
@article{he2025distill,
  title   = {Distill Any Depth: Distillation Creates a Stronger Monocular Depth Estimator},
  author  = {Xiankang He and Dongyan Guo and Hongji Li and Ruibo Li and Ying Cui and Chi Zhang},
  year    = {2025},
  journal = {arXiv preprint arXiv: 2502.19204}
}
```

- **Source Model:** [Distill-Any-Depth by Westlake-AGI-Lab](https://github.com/Westlake-AGI-Lab/Distill-Any-Depth)
- **License:** MIT  
- **Modifications:** Only format conversion (no retraining or weight changes)

> If you use this model, please credit the original authors: Westlake-AGI-Lab.

## 💻 How to Use In VisionDepth3D

Place Folder containing onnx model into weights folder in VisionDepth3D

```
VisionDepth3D¬
              Weights¬
                      Distill Any Depth Small¬
                                            model.onnx
```