bbopen/camo-person-yolo

Camouflaged Person Detector (YOLO, single class)

• Single class: person
• Phase B fine-tuned model on camo fill/background pairs + negatives
• Artifacts: camo-person-yolo.pt (PyTorch), camo-person-yolo.onnx (opset 12, dynamic, simplified), camo-person-yolo.torchscript

Quick usage

Ultralytics (PyTorch)

from ultralytics import YOLO
model = YOLO("bbopen/camo-person-yolo")  # loads camo-person-yolo.pt by default
model.predict(source="image.jpg", imgsz=1280, conf=0.25, iou=0.6)

ONNX Runtime

import onnxruntime as ort, numpy as np, cv2
sess = ort.InferenceSession("camo-person-yolo.onnx", providers=["CUDAExecutionProvider","CPUExecutionProvider"])
im = cv2.imread("image.jpg"); im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
im = cv2.resize(im, (1280,1280)).astype(np.float32)/255.0
im = np.transpose(im,(2,0,1))[None]
outputs = sess.run(None, {"images": im})

Jetson Orin Nano (export to TensorRT)

• Install runtime: python3 -m pip install --upgrade ultralytics
• Export FP16 engine:

yolo export model=camo-person-yolo.pt format=engine half=True imgsz=1280 device=0

• Inference:

yolo task=detect mode=predict model=best_fp16_1280.engine source=path/to/images conf=0.25 iou=0.6 imgsz=1280

Repro/configs

• Optional training args: args.yaml
• Optional dataset layout reference: data.yaml