QuickDraw

Quick, Draw! is a web-based game and accompanying dataset created by Google as part of its AI Experiments program. In the game, players are shown a word and asked to draw it within 20 seconds while the system attempts to guess the drawing. The project aims to explore the diversity of human sketches and to collect data for machine learning research.

The Quick, Draw! dataset, released in 2017, contains tens of millions of doodles across hundreds of categories. It is one of the largest public doodle datasets used for training and evaluating sketch recognition models. Each drawing is recorded as vector strokes—sequences of x and y coordinates with markers separating strokes—and is also provided as a rasterized 28 by 28 grayscale image. The data is typically released for non-commercial research and education purposes, with attention to privacy and responsible use.

Applications of the dataset include training sketch-based classifiers, developing generative and conditional models, and supporting research in few-shot and one-shot learning. Researchers use the data to study how neural networks interpret abstract, informal drawings and to benchmark recognition performance on sketch-like inputs. The project has influenced the development of AI systems capable of interpreting hand-drawn input and has spurred the creation of additional large-scale doodle datasets.

Limitations and considerations include potential biases from the user base that contributed drawings, as well as the simplified resolution and stylized nature of sketches. While not representative of all drawing styles, the Quick, Draw! dataset remains a widely used resource in AI research and education.