DetAny3D

DetAny3D predicts three-dimensional boxes from a single image.

Tasks: 3d detection
Sizes: h at 896 px
Install: pip install "libreyolo[hf]"
Support tier: Sibling tier, since v1.6.0. A separate product surface with its own factory and contract.
Upstream: DetAny3D by OpenDriveLab, Apache-2.0 code; no separate weights license declared by the publisher. Paper, source
Licenses: Code MIT, weights No separate license declared by the publisher. Commercial use

Install

bash

pip install "libreyolo[hf]"

Predict

Python

from libreyolo import LibreDetAny3D, SAMPLE_IMAGE # Requires the separately installed upstream runtime.model = LibreDetAny3D(device="cpu")result = model.predict(SAMPLE_IMAGE, text="person")print(result.boxes3d)

DetAny3D estimates camera intrinsics internally and accepts text, box or point prompts. Install its separate upstream runtime and its grounding dependencies. Default confidence is 0.37 and the text threshold is 0.25.

result.boxes3d holds centers, dimensions and wxyz quaternions in camera coordinates, plus confidence, class and intrinsics. Plotting projects cuboids into the image. Training, validation, tracking and export are not supported. See 3D detection.

Licensing

Check the license on the Hugging Face repository of the specific weights you download. Every checkpoint in the LibreYOLO org carries one, and they are not always the same across a family. That repository is the authoritative source; the summary below describes what applied when this page was last verified.

This is a description of the licenses involved, not legal advice. If the answer matters commercially, read the licenses yourself and take your own counsel.

Original work: DetAny3D, OpenDriveLab
Upstream license: Apache-2.0 code; no separate weights license declared by the publisher
Upstream source: github.com/OpenDriveLab/DetAny3D
LibreYOLO code: MIT
Weights: No separate license declared by the publisher, republished at huggingface.co/LibreYOLO
Interpretation: OpenDriveLab releases the project under Apache-2.0 and declares no separate license for the checkpoint. LibreYOLO mirrors the checkpoint byte-for-byte and adds no terms of its own.

Citation

@article{zhang2025detect,
  title={Detect Anything 3D in the Wild},
  author={Zhang, Hanxue and Jiang, Haoran and Yao, Qingsong and Sun, Yanan and Zhang, Renrui and Zhao, Hao and Li, Hongyang and Zhu, Hongzi and Yang, Zetong},
  journal={arXiv preprint arXiv:2504.07958},
  year={2025}
}

Copied from the authors' citation block at raw.githubusercontent.com/OpenDriveLab/DetAny3D/main/README.md.