pictograph/a2d2

View on Pictograph · Pictograph Research · Creative Commons Attribution-NoDerivatives 4.0

About

A2D2 is a computer-vision dataset from A2D2, hosted on Pictograph. Instance segmentation of road users from the Audi Autonomous Driving Dataset (A2D2): 38,554 frames from the six cameras of an Audi Q7 e-tron, with every car, pedestrian, truck, bicycle, small vehicle, utility vehicle and tractor outlined as a polygon. The most common detected objects are road, car, crosswalk, truck, sidewalk, traffic light. On Pictograph you can browse every annotated image, import it into your own workspace in one click, export it in a dozen formats, or train a model on it directly.

At a glance

Metric	Value
Images	38,554
Annotations	549,965
Classes	7
Annotation types	polygon
Splits	train

Quick start

Load it in one line with the datasets library, then read each record's boxes and class names:

from datasets import load_dataset
from PIL import ImageDraw

ds = load_dataset("pictograph/a2d2", split="train")
example = ds[0]
image = example["image"]            # a PIL image
objects = example["objects"]        # {bbox, categories, category_names}

# draw every bounding box with its class name
draw = ImageDraw.Draw(image)
for (x, y, w, h), name in zip(objects["bbox"], objects["category_names"]):
    draw.rectangle([x, y, x + w, y + h], outline="red", width=3)
    draw.text((x, y - 12), name, fill="red")

# polygon masks live under example["segmentation"]:
#   [{"label": name, "category": idx, "points": [[x, y], ...]}, ...]
image.show()

Prefer a full annotation editor, one-click import, multi-format export, and one-click training? Open this dataset on Pictograph.

Dataset structure

This dataset uses the Hugging Face imagefolder layout: each split directory holds the images plus a metadata.jsonl that links every image to its annotations by file_name.

Field	Description
file_name	Path to the image within the split directory.
objects.bbox	Bounding boxes as [x, y, width, height] (pixels).
objects.categories	Integer class index per box (matches the class list below).
objects.category_names	Human class name per box.
segmentation	List of {label, category, points}; points is a polygon ring [[x, y], ...].

Data instance

One record (bounding boxes are [x, y, width, height] in pixels; the class index maps into the class list below):

{
  "image": <PIL.Image (RGB)>,
  "objects": {
    "bbox": [[172.0, 192.0, 249.4, 152.7]],
    "categories": [0],
    "category_names": ["car"]
  },
  "segmentation": [
    {"label": "car", "category": 0, "points": [[176.9, 207.2], [259.0, 274.1], ...]}
  ]
}

Classes

Class index matches objects.categories in metadata.jsonl.

#	Class	Annotations
0	car	372,389
1	pedestrian	58,401
2	truck	67,059
3	small vehicle	7,812
4	utility vehicle	4,057
5	bicycle	40,136
6	tractor	111

License

Released under Creative Commons Attribution-NoDerivatives 4.0. When you use this data, please credit Audi AG (Geyer et al., 2020).

Source and attribution

The images and annotations come from A2D2, created by Audi AG (Geyer et al., 2020) and released under Creative Commons Attribution-NoDerivatives 4.0. We are grateful to the original authors. If you use this data, please cite the original source above.

Citation

@article{geyer2020a2d2,
  title = {{A2D2}: {Audi Autonomous Driving Dataset}},
  author = {Geyer, Jakob and Kassahun, Yohannes and Mahmudi, Mentar and Ricou, Xavier and Durgesh, Rupesh and Chung, Andrew S. and Hauswald, Lorenz and Pham, Viet Hoang and M{\"u}hlegg, Maximilian and Dorn, Sebastian and Fernandez, Tiffany and J{\"a}nicke, Martin and Mirashi, Sudesh and Savani, Chiragkumar and Sturm, Martin and Vorobiov, Oleksandr and Oelker, Martin and Garreis, Sebastian and Schuberth, Peter},
  journal = {arXiv preprint arXiv:2004.06320},
  year = {2020}
}

---

Published from Pictograph - annotate, train, and deploy from one API.