# Frame Evidence:A Benchmark for Reference Frames and Visual Evidence in Multi-Image Spatial Reasoning

arXiv: https://arxiv.org/abs/2610.19564 · HTML: https://mlsanity.com/paper/2610.19564/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 19 of its 26 references that are in the map.
- First posted on arXiv 14 Oct 2026.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)

## Builds on (most cited first, 19 of 19)

- [SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities](https://mlsanity.com/paper/2401.12168/) (2024)
- [Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces](https://mlsanity.com/paper/2412.14171/) (2024)
- [ERASER: A Benchmark to Evaluate Rationalized NLP Models](https://mlsanity.com/paper/1911.03429/) (2019)
- [Winoground: Probing Vision and Language Models for Visio-Linguistic Compositionality](https://mlsanity.com/paper/2204.03162/) (2022)
- [MuirBench: A Comprehensive Benchmark for Robust Multi-image Understanding](https://mlsanity.com/paper/2406.09411/) (2024)
- [MMSI-Bench: A Benchmark for Multi-Image Spatial Intelligence](https://mlsanity.com/paper/2505.23764/) (2025)
- [3DSRBench: A Comprehensive 3D Spatial Reasoning Benchmark](https://mlsanity.com/paper/2412.07825/) (2024)
- [Do Vision-Language Models Represent Space and How? Evaluating Spatial Frame of Reference Under Ambiguities](https://mlsanity.com/paper/2410.17385/) (2024)
- [Benchmarking Multi-Image Understanding in Vision and Language Models: Perception, Knowledge, Reasoning, and Multi-Hop Reasoning](https://mlsanity.com/paper/2406.12742/) (2024)
- [ReMI: A Dataset for Reasoning with Multiple Images](https://mlsanity.com/paper/2406.09175/) (2024)
- [CityCube: Benchmarking Cross-view Spatial Reasoning on Vision-Language Models in Urban Environments](https://mlsanity.com/paper/2601.14339/) (2026)
- [FoREST: Frame of Reference Evaluation in Spatial Reasoning Tasks](https://mlsanity.com/paper/2502.17775/) (2025)
- [More Images, More Problems? A Controlled Analysis of VLM Failure Modes](https://mlsanity.com/paper/2601.07812/) (2026)
- [Where Does the Answer Come From? Benchmarking View-Level Visual Evidence Identification in Multi-View MLLMs for Autonomous Driving](https://mlsanity.com/paper/2606.09644/) (2026)
- [MindEdit-Bench: Benchmarking Object-Level Counterfactual Spatial Reasoning in VLMs from In-the-Wild Photos](https://mlsanity.com/paper/2607.00491/) (2026)
- [MultiView-Bench: A Diagnostic Benchmark for World-Centric Multi-View Integration in VLMs](https://mlsanity.com/paper/2607.08970/) (2026)
- [CrossView Suite: Harnessing Cross-view Spatial Intelligence of MLLMs with Dataset, Model and Benchmark](https://mlsanity.com/paper/2605.18621/) (2026)
- [SSMNBench: Diagnosing Image-based Cross-View Human-Object Understanding via Single-View Sufficiency and Multi-View Necessity](https://mlsanity.com/paper/2606.25634/) (2026)
- [Learning Multi-View Spatial Reasoning from Cross-View Relations](https://mlsanity.com/paper/2603.27967/) (2026)

## Abstract (arXiv)

A spatial answer depends on the reference frame specified in language and its supporting visual evidence. Item accuracy alone obscures whether correctness persists when either changes. We introduce Frame Evidence, a benchmark of 320 matched triplets and 960 Chinese multiple-choice questions from captured and rendered scenes. One family compares three grounded reference frames over fixed images. The other pairs image deletion annotated as insufficient with answer-preserving addition. Joint metrics require correct answers across all three frames and paired evidence variants. Across 20 configurations on 288 held-out questions, GPT three-shot reaches 50.00% item accuracy, but the best three-frame group accuracy is 5/48 (10.42%), reached by GPT two-shot. Few-shot prompting and task-specific adaptation improve item accuracy without consistently improving joint correctness. Input controls show where images support matched success, while repeated executions reveal answer instability beneath similar scores.