depth-anything-v2-metric-onnx

Depth-Anything-V2-Metric-Hypersim-Small — ONNX

A self-exported ONNX build of depth-anything/Depth-Anything-V2-Metric-Hypersim-Small (indoor metric depth, max_depth = 20 m). Exported so it can run on onnxruntime (CPU/Node) without a PyTorch dependency.

License & attribution

• License: Apache-2.0 — inherited from the base checkpoint and the official Depth-Anything-V2 repository code.
• This repo redistributes only a derived ONNX export of the Apache-2.0 weights, with attribution as required. No other datasets or assets are included.

Export details

• Source: official depth_anything_v2_metric_hypersim_vits.pth.
• torch.onnx.export, opset 17, static input 1 × 3 × 518 × 518, constant-folded.
• Verified against the PyTorch model: max-abs output diff < 1e-3.

Input / output

• Input pixel_values: 1 × 3 × 518 × 518, RGB, ImageNet-normalized (mean [0.485, 0.456, 0.406], std [0.229, 0.224, 0.225]).
• Output predicted_depth: 1 × 518 × 518, metric depth in metres, range (0, 20); larger = farther.

Usage (onnxruntime)

import onnxruntime as ort, numpy as np
sess = ort.InferenceSession("model.onnx", providers=["CPUExecutionProvider"])
depth = sess.run(None, {"pixel_values": x})[0]  # x: 1x3x518x518 float32

Citation

@article{depth_anything_v2,
  title={Depth Anything V2},
  author={Yang, Lihe and Kang, Bingyi and Huang, Zilong and Zhao, Zhen and Xu, Xiaogang and Feng, Jiashi and Zhao, Hengshuang},
  journal={arXiv:2406.09414},
  year={2024}
}