VisualOverload is a scene image understanding evaluation dataset that aims to examine the model's visual understanding and reasoning ability of details in complex scenes without relying on external knowledge. This dataset contains 2,720 question-answer pairs, consisting of public-domain, high-resolution paintings that often feature multiple characters, actions, subplots, and complex backgrounds. The questions are manually designed to comprehensively test the model's scene understanding. This dataset is suitable for visual question answering research, detailed image understanding and reasoning, and evaluation of complex scenes with multiple characters and elements.

Citation

@InProceedings{Gavrikov_2026_visualoverload, author={Paul Gavrikov and Wei Lin and M. Jehanzeb Mirza and Soumya Jahagirdar and Muhammad Huzaifa and Sivan Doveh and Serena Yeung-Levy and James Glass and Hilde Kuehne}, title={{VisualOverload}: Probing Visual Understanding of VLMs in Really Dense Scenes}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, month = {June}, year = {2026} }