Preprint
Aug 2026
SpatioLM: Towards General Physical Spatial Intelligence in Vision-Language Models
This work designs a plug-and-play and non-invasive spatio-vision module that elicits the spatial knowledge inherent in VLMs and innovatively leverage pseudo depth and camera information as supervision to guide the model in learning physically coherent representations.
Jing Wu, Jianhua Wu, Jiayi Guan et al.
· 0 citations