beaunix/colcar-vision

ColCar Vision — Vehicle Inspection Models

This repository hosts the three vision models that power ColCar (Car-Inspector), an agent that assists with vehicle inspections, feeding a Colombian RTM compliance RAG and pricing engine.

All three models are built with Ultralytics YOLO (YOLOv11 and YOLOv8). Full credit to Ultralytics for the open YOLO architecture and training tooling.

Models

1. Damage Segmentation (YOLOv11-medium, instance segmentation)

colcar-damage-seg.{pt,onnx}

• Classes: crack, scratch, dent, glass_shatter, lamp_broken, tire_flat
• Training data: Car-Damage-Type-Detection-End-Game, Roboflow Universe, CC BY 4.0

2. Car Parts Segmentation (YOLOv11-medium, instance segmentation)

colcar-parts-seg.{pt,onnx}

• Classes (21): wheel, hood, trunk, bumpers, doors, glass, lights, mirrors, tailgate, etc.
• Training data: car-seg from https://docs.ultralytics.com/datasets/segment/carparts-seg by Gianmarco Russo, Roboflow Universe, CC BY 4.0

3. Tyre Defect Detection (YOLOv8-medium, object detection)

colcar-tyre-defect-det.{pt,onnx}

• Classes: Good, Bulge, Cracks, Flat spots, Pitting, Puncture
• Training data: Tyre defect detection, Roboflow Universe, CC BY 4.0

Usage

ONNX Runtime (inference only)

```python import onnxruntime as ort

session = ort.InferenceSession("colcar-damage-seg.onnx") outputs = session.run(None, {"images": input_tensor}) ```

Ultralytics (.pt — inference or further training)

```python from ultralytics import YOLO

model = YOLO("colcar-damage-seg.pt") results = model.predict("car_photo.jpg") ```

License

Released under AGPL-3.0, in line with the Ultralytics YOLO license these models were trained with (no Enterprise license was purchased).

Related

• Project repo: https://github.com/BeauBryanDev/colcar