{"paper":{"title":"Beyond Localization: A Comprehensive Diagnosis of Perspective-Conditioned Spatial Reasoning in MLLMs from Omnidirectional Images","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Multimodal large language models show a large gap between perception and perspective-conditioned spatial reasoning on omnidirectional images.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"(2) Guangzhou University, (3) Queen Mary University of London, 4) ((1) The Hong Kong Polytechnic University, (4) HKUST (Guangzhou)), Ioannis Patras, Jiaxing Li, Wai Keung Wong, Xu Zheng, Yuangong Chen","submitted_at":"2026-05-12T17:11:17Z","abstract_excerpt":"Multimodal Large Language Models (MLLMs) show strong visual perception, yet remain limited in reasoning about space under changing viewpoints. We study this challenge as Perspective-Conditioned Spatial Reasoning (PCSR) in 360-degree omnidirectional images, where broad scene coverage reduces ambiguity from partial observations without eliminating the need for viewpoint-dependent inference. To assess this capability, we introduce PCSR-Bench, a diagnostic benchmark of 84,373 question-answer pairs from 2,600 omnidirectional images across 26 indoor environments. PCSR-Bench contains eight tasks span"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"PCSR is a key bottleneck in current MLLMs and highlight limited but meaningful room for recovery under targeted optimization.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The eight tasks in PCSR-Bench accurately isolate perspective-conditioned spatial reasoning without confounding effects from omnidirectional projection artifacts or question-generation biases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A new benchmark reveals MLLMs achieve only 13% or lower accuracy on advanced perspective-conditioned spatial tasks in omnidirectional images, with RL reward shaping raising a 7B model from 31% to 60% in controlled settings.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Multimodal large language models show a large gap between perception and perspective-conditioned spatial reasoning on omnidirectional images.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7a17d1ad70428c0b56da8d7e0427ad45f097508494f042233811c88d771824d6"},"source":{"id":"2605.12413","kind":"arxiv","version":3},"verdict":{"id":"5063b759-8873-4a38-b55c-4318642c8606","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T21:34:00.189518Z","strongest_claim":"PCSR is a key bottleneck in current MLLMs and highlight limited but meaningful room for recovery under targeted optimization.","one_line_summary":"A new benchmark reveals MLLMs achieve only 13% or lower accuracy on advanced perspective-conditioned spatial tasks in omnidirectional images, with RL reward shaping raising a 7B model from 31% to 60% in controlled settings.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The eight tasks in PCSR-Bench accurately isolate perspective-conditioned spatial reasoning without confounding effects from omnidirectional projection artifacts or question-generation biases.","pith_extraction_headline":"Multimodal large language models show a large gap between perception and perspective-conditioned spatial reasoning on omnidirectional images."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.12413/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-19T22:21:57.860991Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T10:35:35.000735Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T08:01:18.507009Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T07:31:32.721814Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"13154b31e2f6361e4ea177480e6d28d63cd716ad5c838290c8c900c4461b2fe7"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"3ea6f60a156eda1df9f77acbb6fef223fedf3167d30d1e6554a91ddc180e44ec"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}