---
license: unknown
language:
- en
pretty_name: Car Part Damage (Parts)
size_categories:
- 1K<n<10K
task_categories:
- image-segmentation
- object-detection
tags:
- car
- car-parts
- supervised
---

# Car Part Damage (Parts)

## Data format
Each example contains:
- `image`: RGB image (Hugging Face `Image` feature; embedded so the viewer works)
- `image_id`: original filename
- `width`: image width in pixels
- `height`: image height in pixels
- `annotations`: list of detection objects (COCO-like), each with:
  - `id`: integer object id from the annotation file
  - `category`: car part label (string)
  - `bbox`: **[x_min, y_min, x_max, y_max]** in absolute pixel coordinates (xyxy)
  - `segmentation`: list of polygons, each polygon is a flat list
    `[x1, y1, x2, y2, ...]` in pixel coordinates.
    `segmentation[0]` is the exterior polygon; subsequent entries are holes if present.

Bounding boxes are derived from the exterior polygon by taking min/max of x and y.

## Splits
- `train`: 948 images
- `test`: 50 images

## Labels
The part classes are:
Back-bumper, Back-door, Back-wheel, Back-window, Back-windshield, Fender, Front-bumper, Front-door,
Front-wheel, Front-window, Grille, Headlight, Hood, License-plate, Mirror, Quarter-panel,
Rocker-panel, Roof, Tail-light, Trunk, Windshield.

## Usage
```python
from datasets import load_dataset

ds = load_dataset("moondream/car_part_damage", split="train")
print(ds[0]["annotations"][0]["bbox"])  # [x_min, y_min, x_max, y_max]
print(ds[0]["annotations"][0]["category"])  # part name
```