Depth Anything V2 Small
onnx-community/depth-anything-v2-small
published Jun 2024 · updated Apr 2026
Depth Anything V2 Small is a depth estimation model that predicts dense depth maps from a single image.
specs
| Task | Depth Estimation |
| Architecture | Vision Transformer-based |
| Parameters | 24.8M |
| License | Apache-2.0 |
about this model
depth-anything-v2-small is a depth estimation model that predicts dense depth maps from a single image. It is a version of the Depth Anything V2 architecture, optimized with ONNX weights for efficient inference.
The model produces fine-grained depth details that outperform Depth Anything V1 while being more robust than V1 and Stable Diffusion-based approaches such as Marigold and Geowizard. It is roughly 10 times faster and more lightweight than Stable Diffusion-based depth models. Depth-Anything-V2-Small has 24.8 million parameters and was trained on 595,000 synthetic labeled images combined with over 62 million real unlabeled images. The work was accepted at NeurIPS 2024.
Metric depth models derived from Depth-Anything-V2-Small and Base have also been released, extending its utility to applications requiring absolute depth values. The model is licensed under Apache-2.0.
Gigarouter hosts depth-anything-v2-small as a managed, OpenAI-compatible API. Developers can integrate depth estimation into their applications without managing infrastructure.
best for
- ·Monocular depth estimation for single images
- ·3D scene understanding and reconstruction
- ·Augmented reality occlusion handling
FAQ
It is best for efficient monocular depth estimation, producing dense depth maps from a single image with high speed and low compute.
At 24.8M parameters, it is lightweight and about 10x faster than Stable Diffusion-based depth models while offering more fine-grained details than its predecessor.
It is released under the Apache-2.0 license, allowing commercial use with attribution.
Input: a single image (RGB). Output: a depth map of the same resolution, typically as a single-channel grayscale image or tensor.
Use the OpenAI-compatible endpoint with your API key, passing an image URL or base64-encoded image, and the response will contain the depth map.
We're benchmarking and onboarding Depth Anything V2 Small as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.