Spatial-Conditioned Reasoning in Long-Egocentric Videos
This work addresses spatial reasoning limitations in long egocentric videos for navigation applications, but it is incremental as it focuses on input-level modifications without new model architectures.
The paper tackled the challenge of spatial reasoning in long-horizon egocentric videos by studying how explicit spatial signals like depth maps influence vision-language models without architectural changes, finding that depth-aware representations improve performance on safety-critical tasks such as pedestrian and obstruction detection.
Long-horizon egocentric video presents significant challenges for visual navigation due to viewpoint drift and the absence of persistent geometric context. Although recent vision-language models perform well on image and short-video reasoning, their spatial reasoning capability in long egocentric sequences remains limited. In this work, we study how explicit spatial signals influence VLM-based video understanding without modifying model architectures or inference procedures. We introduce Sanpo-D, a fine-grained re-annotation of the Google Sanpo dataset, and benchmark multiple VLMs on navigation-oriented spatial queries. To examine input-level inductive bias, we further fuse depth maps with RGB frames and evaluate their impact on spatial reasoning. Our results reveal a trade-off between general-purpose accuracy and spatial specialization, showing that depth-aware and spatially grounded representations can improve performance on safety-critical tasks such as pedestrian and obstruction detection.