occurra/object_detection

ONNX exports of Ultralytics YOLOv11 in the configurations the occurra object_detection agent ships with.

Two model sizes (nano n, small s), four precision variants, and two input resolutions. All files are self-contained (no external-data sidecars).

Filename convention

yolo11{n,s}_{apple,fp16,fp8,int8}_{640x640,640x480}.onnx

Token	Meaning
n / s	YOLOv11 nano (5 MB) or small (19 MB)
apple	FP16, NMS-free, batch=1, static — CoreML / Apple ANE friendly. uint8 input.
fp16	FP16 weights, NMS embedded. Default for NVIDIA TensorRT EP.
fp8	FP8 quantized via TensorRT QDQ. Smallest VRAM footprint on Blackwell / Hopper.
int8	INT8 quantized with QDQ nodes embedded in the graph. No sidecar calibration cache needed.
640x640	Square input — used on Apple (apple variant) and as the upstream default.
640x480	4:3 input — ~25% fewer FLOPs than 640×640 on cameras with 4:3 aspect, measurably faster on NVIDIA TensorRT.

The object_detection agent reads the input shape directly from the loaded ONNX (graph.input[0].type) — no sidecar config, the file name is informational.

Which file to pick

Hardware	Recommended
Apple Silicon (CoreML / ANE)	yolo11n_apple_640x640.onnx
NVIDIA RTX 4000+ / Blackwell	yolo11n_fp8_640x480.onnx
NVIDIA older (no FP8)	yolo11n_int8_640x480.onnx or yolo11n_fp16_640x480.onnx
Higher accuracy (any NVIDIA)	swap the n for s (3–4× slower, marginally better mAP)

Source

Trained Ultralytics checkpoints (yolo11n.pt, yolo11s.pt) are downloaded from Ultralytics' release feed and re-exported via the occurra toolbox's ai_agent_toolbox/agents/python/object_detection/scripts/main.py (NMS-free for Apple, with-NMS for NVIDIA; FP8/INT8 use TensorRT QDQ).

License

The model weights inherit Ultralytics YOLOv11's AGPL-3.0 license. Commercial use requires a separate enterprise license from Ultralytics — the ONNX export does not change that.