Instructions to use en970/depth-anything-v3-small-onnx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers.js
How to use en970/depth-anything-v3-small-onnx with Transformers.js:
// npm i @huggingface/transformers import { pipeline } from '@huggingface/transformers'; // Allocate pipeline const pipe = await pipeline('depth-estimation', 'en970/depth-anything-v3-small-onnx');
Depth Anything V3 Small — 8-bit ONNX
An 8-bit dynamically quantised export of onnx-community/depth-anything-v3-small, produced for depth-realtime, a browser-based real-time depth viewer.
The upstream repository ships float32 only, at 105 MB, which is more than a web page should ask a visitor to download before it can show anything. This export is 28.9 MB.
Fidelity
Measured against the float32 original on three scenes, at 322 px input:
| Scene | Pearson r | float32 time | 8-bit time |
|---|---|---|---|
| Portrait, deep background | 0.9977 | 218 ms | 101 ms |
| Indoor, mid range | 0.9902 | 171 ms | 113 ms |
| Indoor, close range | 0.9984 | 127 ms | 90 ms |
Distinct depth levels surviving in the far third of the frame, after the percentile normalisation and 8-bit transport a real viewer applies: 179 → 177, 84 → 91, 50 → 53. The quantisation does not cost background structure.
Licence
Apache-2.0, inherited from the base model.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for en970/depth-anything-v3-small-onnx
Base model
onnx-community/depth-anything-v3-small