FCOS3D

FCOS3D predicts three-dimensional boxes from a single image.

Tasks: 3d detection
Sizes: r101 at 1600 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: FCOS3D by OpenMMLab, Apache-2.0 code. Paper, source
Licenses: Code MIT, weights No separate artifact grant recorded. Commercial use

Install

bash

pip install "libreyolo[hf]"

Predict

Python

from libreyolo import LibreFCOS3D, SAMPLE_IMAGEfrom PIL import Image model = LibreFCOS3D(device="cpu")# Rough pinhole guess so the snippet runs. For real metric boxes, pass# your camera's measured 3x3 matrix for the original image size.w, h = Image.open(SAMPLE_IMAGE).sizef = float(max(w, h))intrinsics = [[f, 0, w / 2], [0, f, h / 2], [0, 0, 1]]result = model.predict(SAMPLE_IMAGE, intrinsics=intrinsics)print(result.boxes3d)

FCOS3D runs natively on CPU or CUDA and requires original-image camera intrinsics. MPS is not supported. The default confidence is 0.05, rotated NMS IoU is 0.8 and max_det is 200.

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: FCOS3D, OpenMMLab
Upstream license: Apache-2.0 code
Upstream source: github.com/open-mmlab/mmdetection3d
LibreYOLO code: MIT
Weights: No separate artifact grant recorded, republished at huggingface.co/LibreYOLO
Interpretation: The upstream repository declares Apache-2.0. A distinct per-artifact checkpoint license is not recorded in the library source.

Citation

@inproceedings{wang2021fcos3d,
	title={{FCOS3D: Fully} Convolutional One-Stage Monocular 3D Object Detection},
	author={Wang, Tai and Zhu, Xinge and Pang, Jiangmiao and Lin, Dahua},
	booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
	year={2021}
}

Copied from the authors' citation block at raw.githubusercontent.com/open-mmlab/mmdetection3d/main/configs/fcos3d/README.md.