libreyolo fcos3d
Run fcos3d 3D detection from the command line.
Synopsis
bash
libreyolo fcos3d source=IMAGE intrinsics=camera.npy [OPTIONS]Arguments
| Argument | Default | Meaning |
|---|---|---|
source | required | Image path or directory |
model | upstream default checkpoint | Local checkpoint or the mirrored default filename |
intrinsics | required | Original-image 3x3 camera calibration .npy file |
device | auto | auto, cpu, or a CUDA device |
conf | None | Joint confidence threshold |
iou | None | Rotated bird's-eye-view NMS threshold |
max-det | 200 | Maximum number of detections per image |
save | False | Save projected cuboids |
output-path | None | Single-image output filename |
json | false | JSON output to stdout |
quiet | False | Suppress stderr |
--help-json | flag | Dump command schema as JSON |
Examples
Inspect the installed command schema:
bash
libreyolo fcos3d --help-jsonPredict on the shipped sample image. For commands requiring calibration,
camera.npy must contain its measured original-image 3x3 intrinsic matrix.
bash
libreyolo fcos3d source=https://raw.githubusercontent.com/LibreYOLO/libreyolo/release/libreyolo/assets/parkour.jpg intrinsics=camera.npyRequest machine-readable results:
bash
libreyolo fcos3d source=https://raw.githubusercontent.com/LibreYOLO/libreyolo/release/libreyolo/assets/parkour.jpg intrinsics=camera.npy --json --quietNotes
Configuration, model-loading and I/O errors exit through the shared CLI
error handler. --json writes structured results to stdout; diagnostics use
stderr. --quiet suppresses stderr. See fcos3d
for installation, calibration and model constraints.