mlsanityData to 14 Oct 2026

Lost in Space? Vision-Language Models Struggle with Relative Camera Pose Estimation

arXiv 2601.22228 ↗First posted 29 Jan 2026cs.CV cs.AI cs.CLcited by 2

On the map: LLM reasoning → Spatial and embodied reasoning

also in 1 more
Abstract (arXiv)

We study whether vision-language models (VLMs) can solve relative camera pose estimation (RCPE) from image pairs, a direct test of multi-view spatial reasoning. We cast RCPE as a discrete verbal classification task and introduce textttVRRPI-Bench, built from real RGB-D frames with object-centric camera motion, and textttVRRPI-Diag, which isolates individual motion degrees of freedom. Humans (0.91) and specialized geometric pipelines such as LoFTR (0.99) solve the task reliably, yet the best VLM reaches only 0.66 and most others remain near random. Our analyses show that this gap is not basic spatial competence: strong VLMs are near ceiling on single-image benchmarks, but most remain near random once reasoning must span views. They are unstable under source-target reversal (best 59.7% consistency) and remain weak even in simplified single-DoF settings, especially on optical-axis motions such as roll and depth translation (GPT-5: 0.46 on roll). These failures are useful: they localize concrete missing capabilities, namely cross-view correspondence, view-consistent reasoning, and projective camera-motion understanding, making RCPE a targeted diagnostic for improving multi-view spatial reasoning in VLMs.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Spatial and embodied reasoning, one of 17 directions in LLM reasoning. Builds on 24 of its 42 references that are in the map. First posted on arXiv 29 Jan 2026.

Builds on

24 of 42 references.

Closest ancestors · 8

Same direction, newest first

  1. 2025Reinforcing Spatial Reasoning in Vision-Language Models with Interwoven Thinking and Visual Drawing11 Jun
  2. 2025SSR: Enhancing Depth Perception in Vision-Language Models via Rationale-Guided Spatial Reasoning18 May
  3. 2025SpatialCoT: Advancing Spatial Reasoning through Coordinate Alignment and Chain-of-Thought for Embodied Task Planning17 Jan
  4. 2024Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces18 Dec
  5. 2024Sparkle: Mastering Basic Spatial Capabilities in Vision Language Models Elicits Generalization to Spatial Reasoning21 Oct
  6. 2024SpatialPIN: Enhancing Spatial Reasoning Capabilities of Vision-Language Models through Prompting and Interacting 3D Priors18 Mar
  7. 2024SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities22 Jan
  8. 2023What's "up" with vision-language models? Investigating their struggle with spatial reasoning30 Oct

Other references · 16

  1. 2017ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenescited by 3,117
  2. 2024LLaVA-OneVision: Easy Visual Task Transfercited by 2,719
  3. 2021LoFTR: Detector-Free Local Feature Matching with Transformerscited by 984
  4. 2023ScanNet++: A High-Fidelity Dataset of 3D Indoor Scenescited by 764
  5. 2023Large Language Models Are Not Robust Multiple Choice Selectorscited by 486
  6. 2025GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learningcited by 466
  7. 2022LM-Nav: Robotic Navigation with Large Pre-Trained Models of Language, Vision, and Actioncited by 439
  8. 2025Spatial-MLLM: Boosting MLLM Capabilities in Visual-based Spatial Intelligencecited by 217
  9. 2024RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Roboticscited by 136
  10. 2025MM-Spatial: Exploring 3D Spatial Understanding in Multimodal LLMscited by 73
  11. 2024Isambard-AI: a leadership class supercomputer optimised specifically for Artificial Intelligencecited by 62
  12. 2023Large Language Models are Visual Reasoning Coordinatorscited by 58
4 more
  1. 2022The 8-Point Algorithm as an Inductive Bias for Relative Pose Prediction by ViTscited by 38
  2. 2023Are Large Language Models Temporally Grounded?cited by 27
  3. 2022Leveraging Image Matching Toward End-to-End Relative Camera Pose Regressioncited by 6
  4. 2025LRR-Bench: Left, Right or Rotate? Vision-Language models Still Struggle With Spatial Understanding Taskscited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Planning with the Viewscited by 3
  2. 2026Agentic Relative Camera Pose Estimation via Learned Ranking and Verificationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026The Spatial Blindspot of Vision-Language Models
  2. 2025Beyond Recognition: Evaluating Visual Perspective Taking in Vision Language Models
  3. 2026Geometric Encoding for Spatial Reasoning in Vision-Language Models
  4. 2026Predicting Camera Pose from Perspective Descriptions for Spatial Reasoning
  5. 2025Vision language models are unreliable at trivial spatial cognition
  6. 2025ViewSpatial-Bench: Evaluating Multi-perspective Spatial Localization in Vision-Language Models
  7. 2025SpaRE: Enhancing Spatial Reasoning in Vision-Language Models with Synthetic Data
  8. 2024UniBench: Visual Reasoning Requires Rethinking Vision-Language Beyond Scaling