Computer vision as one of the major sub-domains in AI has demonstrated unprecedented growth in the past decade both academically and commercially. Getting access to high quality datasets is critical to conduct research or build products. Thanks to the vibrant community, more and more datasets are made publicly available. However, due to the lack of standard and unified distribution channels, it is difficult to find the datasets of interest easily as well as keeping up-to-date with the new release.

VisualData Discovery is created to change this. It is part of the overall VisualData platform with a mission to allow people working in the computer vision field to easily discover and share latest datasets with the community.

VisualData Discovery automatically monitors various sources, e.g. arXiv, university labs, social media to track the release of new datasets and extracts critical information from the source page. Manual review is followed to verify the information before publishing.

VisualData Discovery is unique in the following key aspects:

- Specializes in computer vision datasets and is one of the largest dataset collections publicly available.

- Does not require specific metadata format for the dataset page to be indexed, giving creators most flexibility to host their datasets.

- All datasets are categorized into fine-grained topics to help search.

- Each dataset includes rich information like models, code, license, publications and more.

- Designed as ‘Instagram for datasets’ which features cover images and continuous scrolling, making the experience inspiring and fun.

Spiros Sirmakessis

University of Peloponnese