Sherlock

The Sherlock dataset contains 103K images collected from the Visual Genome and Visual Common Sense Reasoning datasets. These images are split into 90K training, 6.6K validation, and 6.6K testing sets. Each image is re-annotated with an average of 3.5 observation-inference pairs, forming 363K samples.

Data and Resources

Cite this as

Hao Zhang, Yeo Keat Ee, Basura Fernando (2024). Dataset: Sherlock. https://doi.org/10.57702/q3p8gtpy

DOI retrieved: December 16, 2024

Additional Info

Field Value
Created December 16, 2024
Last update December 16, 2024
Defined In https://doi.org/10.48550/arXiv.1801.10442
Author Hao Zhang
More Authors
Yeo Keat Ee
Basura Fernando
Homepage https://leaderboard.allenai.org/sherlock/submissions/public