3D object detection

3D detection returns object centers, dimensions and orientations in camera coordinates.

Models

WildDet3D, 3D-MOOD, FCOS3D and DetAny3D return Results.boxes3d. FCOS3D runs natively; the other adapters require their separate upstream runtimes.

Predict

Use the model page's calibrated example. WildDet3D, 3D-MOOD and FCOS3D require a measured 3x3 intrinsic matrix for the original image. DetAny3D estimates intrinsics internally. Box orientations use wxyz quaternions. result.plot() projects cuboids into the source image.

Train

These four adapters do not support training, validation, tracking or export.

Verified against LibreYOLO v1.6.0.