# Third-party notice — Depth Anything (Small)

This directory contains an unmodified pretrained model checkpoint redistributed under its original license.

- **Model**: Depth Anything, Small variant (DINOv2-Small backbone, DPT decoder head) — monocular relative depth estimation
- **Original model**: LiheYoung / Depth Anything team
- **Original repository**: https://github.com/LiheYoung/Depth-Anything
- **HuggingFace model card**: https://huggingface.co/LiheYoung/depth-anything-small-hf
- **ONNX conversion**: Xenova (converted + dynamically quantized for browser / Transformers.js use)
- **Upstream ONNX repository**: https://huggingface.co/Xenova/depth-anything-small-hf
- **License**: Apache License 2.0 (see `LICENSE` in this directory, fetched verbatim from the original GitHub repository)
- **Commercial use**: Permitted. Apache-2.0 has no non-commercial restriction; it only requires retaining the license/copyright notice and stating any changes made to the source.
- **Modifications made**: None. `model.onnx` is `onnx/model_quantized.onnx` from the upstream ONNX repository, used exactly as published (dynamic INT8 quantization applied upstream by Xenova via Optimum, not by this site).
- **Asset source**: https://huggingface.co/Xenova/depth-anything-small-hf/resolve/main/onnx/model_quantized.onnx
- **File size**: 27,524,771 bytes
- **SHA-256**: `e4e2ffe9b0396ebd4a0c09b855762b186aab338e79f76531b09777aafc91db84`

Preprocessing/postprocessing logic used in this site's `depth-estimation.js` was independently
implemented in JavaScript, verified against real image inference (resize so the longer side is
~518px, rounded to a multiple of 14, ImageNet mean/std normalization, min-max normalized
per-image depth colorized with the Turbo colormap) and is original code, not redistributed from
the upstream repository.

The Turbo colormap implementation in `depth-estimation.js` is a JavaScript port of the polynomial
approximation published by Anton Mikhailov / Google
(https://research.google/blog/turbo-an-improved-rainbow-colormap-for-visualization/), released
under Apache License 2.0.