Spatial-SSRL-81k is a self-supervised vision-language dataset for spatial understanding and spatial reasoning, released in 2025 by the Shanghai Artificial Intelligence Laboratory in collaboration with Shanghai Jiao Tong University, the Chinese University of Hong Kong, and other institutions. The related research paper is titled "...".Spatial-SSRL: Enhancing Spatial Understanding via Self-Supervised Reinforcement Learning”The aim is to provide large models with spatial awareness capabilities that do not require manual annotation, thereby improving their reasoning and generalization performance in multimodal scenarios.

This dataset contains 81,053 automatically generated question-and-answer samples, constructed based on COCO RGB images and DIODE and MegaDepth RGB-D images. It covers a variety of question formats, including ranking tasks, multiple-choice questions with image options, and multiple-choice questions with text options, covering diverse indoor and outdoor real-world scenarios.