# Quick, Draw! Dataset: 50 Million Hand-Drawn Sketches in 345 Categories as Stroke Vectors and 28x28 Bitmaps (Google Creative Lab)

## Facts

- Provider: Google Creative Lab
- License: CC-BY-4.0 (https://github.com/googlecreativelab/quickdraw-dataset)
- Stored at: Google Cloud Storage, gs://quickdraw_dataset/full (us)
- Access: public, readable without an account
- Category: Computer Vision
- Source: https://github.com/googlecreativelab/quickdraw-dataset
- Page: https://scigantic.com/archive/67d2110f-56e2-5d7f-b028-450c07a32431
- Keywords: Computer Vision, Google Cloud Storage

## About

The Quick, Draw! Dataset holds 50 million drawings across 345 object categories, drawn by players of the online game Quick, Draw!, who were asked to sketch a named object while a neural network tried to recognize it. Each drawing is a timestamped sequence of pen strokes (vector x, y and time coordinates) with metadata: the word the player was asked to draw, whether the game recognized it, a timestamp and the player's two-letter country code. This row covers the full/ folder, which holds one file per category in four forms: raw/ (moderated ndjson with full timing), simplified/ (ndjson aligned to a 256x256 grid, resampled with 1-pixel spacing and simplified with the Ramer-Douglas-Peucker algorithm, epsilon 2.0), binary/ (a compact .bin encoding of the simplified strokes) and numpy_bitmap/ (28x28 grayscale .npy bitmaps, MNIST-style). The sibling folder gs://quickdraw_dataset/sketchrnn/ in the same bucket holds the .npz stroke-3 files with 70K train, 2.5K validation and 2.5K test drawings per category used to train Sketch-RNN, plus .full.npz sets. Used for sketch recognition, sequence generation, cross-cultural studies of drawing and doodle classification benchmarks. Read the ndjson with any JSON-lines reader and the .npy/.npz files with NumPy (np.load(..., encoding='latin1', allow_pickle=True) for the npz). License: CC BY 4.0, made available by Google. Cite the Quick, Draw! Dataset (Google Creative Lab) and, for the sketchrnn files, Ha and Eck, A Neural Representation of Sketch Drawings (arXiv:1704.03477).

## Load the data

```python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('quickdraw_dataset/full/', detail=False)[:10]
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

Example file: `full/binary/The Eiffel Tower.bin`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/67d2110f-56e2-5d7f-b028-450c07a32431
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/67d2110f-56e2-5d7f-b028-450c07a32431
- Over MCP: `get_schema_card` with id `67d2110f-56e2-5d7f-b028-450c07a32431` at https://api.scigantic.com/mcp

## Related topics

- [Computer Vision datasets](https://scigantic.com/topics/computer-vision)
