---
base_model:
- Ultralytics/YOLO26
datasets:
- mitbersh/car-damage-segmentation-yolo
pipeline_tag: image-segmentation
---
# AutoInspect - Car Damage Segmentation (YOLO26)

Модель для сегментации повреждений автомобиля на изображении.
Часть проекта [**AutoInspect**](https://github.com/DedovInside/AutoInspect/tree/ml/ml).

## Task

Сегментация повреждений автомобиля.

## Overview

Модель построена на **YOLO26-m**. Основные параметры:

- **Input image size:** `896`
- **Number of classes:** `6`

Модель обучалась в [Kaggle-ноутбуке](https://www.kaggle.com/code/brshtskmit/train-car-damage-segmentation-yolov26-m-cardd).

## Quick Start

### Installation

```bash
pip install -U ultralytics huggingface_hub
```

### Python Inference

```python
from huggingface_hub import hf_hub_download
from ultralytics import YOLO

repo_id = "mitbersh/car-damage-segmentation"

weights_path = hf_hub_download(
    repo_id=repo_id,
    filename="damage_segmentation.pt"
)

model = YOLO(weights_path)

results = model.predict(
    source="path/to/car_image.jpg",
    imgsz=896,
    conf=0.25,
    save=True
)

print(results)
```

### Alternative: CLI Inference

```bash
yolo segment predict \
  model=damage_segmentation.pt \
  source=path/to/car_image.jpg \
  imgsz=896 \
  conf=0.25 \
  save=True
```