{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:BG6KOY6CM3RKLSDY6X2RSNMIAC","short_pith_number":"pith:BG6KOY6C","schema_version":"1.0","canonical_sha256":"09bca763c266e2a5c878f5f519358800899b740e3662a488acd43ee337938f74","source":{"kind":"arxiv","id":"2407.06581","version":6},"attestation_state":"computed","paper":{"title":"Vision language models are blind: Failing to translate detailed visual features into words","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.AI","authors_text":"Anh Totti Nguyen, Logan Bolton, Mohammad Reza Taesiri, Pooyan Rahmanzadehgervi","submitted_at":"2024-07-09T06:20:17Z","abstract_excerpt":"While large language models with vision capabilities (VLMs), e.g., GPT-4o and Gemini 1.5 Pro, score high on many vision-understanding benchmarks, they are still struggling with low-level vision tasks that are easy to humans. Specifically, on BlindTest, our suite of 7 very simple tasks, including identifying (a) whether two circles overlap; (b) how many times two lines intersect; (c) which letter is being circled in a word; and (d) the number of circles in an Olympic-like logo, four state-of-the-art VLMs are only 58.07% accurate on average. Claude 3.5 Sonnet performs the best at 77.84% accuracy"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2407.06581","kind":"arxiv","version":6},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-07-09T06:20:17Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"2175894e042d99792aaf61c8ee9d2fa6bb3b8a564ea5b6c8f0053010e56afed7","abstract_canon_sha256":"e06eec9a522049bb54688489248f8922c21121b3f70e5e0429f151449eae9464"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:40:01.797649Z","signature_b64":"5eJNAew1jk7izm6uRF7GBuIqL84H4s5sPIb3gQLZcyx8/j2zEjUu97Bh8+E4LroSt5CL4RKikywWFkcA70yoAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"09bca763c266e2a5c878f5f519358800899b740e3662a488acd43ee337938f74","last_reissued_at":"2026-07-05T10:40:01.797129Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:40:01.797129Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Vision language models are blind: Failing to translate detailed visual features into words","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.AI","authors_text":"Anh Totti Nguyen, Logan Bolton, Mohammad Reza Taesiri, Pooyan Rahmanzadehgervi","submitted_at":"2024-07-09T06:20:17Z","abstract_excerpt":"While large language models with vision capabilities (VLMs), e.g., GPT-4o and Gemini 1.5 Pro, score high on many vision-understanding benchmarks, they are still struggling with low-level vision tasks that are easy to humans. Specifically, on BlindTest, our suite of 7 very simple tasks, including identifying (a) whether two circles overlap; (b) how many times two lines intersect; (c) which letter is being circled in a word; and (d) the number of circles in an Olympic-like logo, four state-of-the-art VLMs are only 58.07% accurate on average. Claude 3.5 Sonnet performs the best at 77.84% accuracy"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.06581","kind":"arxiv","version":6},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2407.06581/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2407.06581","created_at":"2026-07-05T10:40:01.797186+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.06581v6","created_at":"2026-07-05T10:40:01.797186+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.06581","created_at":"2026-07-05T10:40:01.797186+00:00"},{"alias_kind":"pith_short_12","alias_value":"BG6KOY6CM3RK","created_at":"2026-07-05T10:40:01.797186+00:00"},{"alias_kind":"pith_short_16","alias_value":"BG6KOY6CM3RKLSDY","created_at":"2026-07-05T10:40:01.797186+00:00"},{"alias_kind":"pith_short_8","alias_value":"BG6KOY6C","created_at":"2026-07-05T10:40:01.797186+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":21,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25066","citing_title":"Do vision-language models search like humans? Reasoning tokens as a reaction-time analog in classic visual-search paradigms","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26529","citing_title":"The Inattentional Gap: Task-Conditioned Language and Vision Models Omit the Safety-Critical Signals They Can Otherwise Report","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20770","citing_title":"Beyond 'One Language, One Script': Quantifying Orthographic Bias in Multilingual VLMs with PuMVR","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18216","citing_title":"Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients","ref_index":142,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07861","citing_title":"The Last Visible Pixel: Probing Fine-Scale Perception in Vision-Language Models","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00434","citing_title":"Information-Regularized Attention for Visual-Centric Reasoning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07639","citing_title":"MOSS-Video-Preview: Toward Real-Time Video Understanding via Cross-Attention","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20448","citing_title":"Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25427","citing_title":"Binding Visual Features Point by Point","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31148","citing_title":"SpatialAct: Probing Spatial Reasoning-to-Action Capabilities of VLM Agents in 3D Scenes","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00564","citing_title":"Decomposed On-Policy Distillation for Vision-Language Reasoning: Steering Gradients for Visual Grounding","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2503.16549","citing_title":"MathFlow: Enhancing the Perceptual Flow of MLLMs for Visual Mathematical Problems","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2505.23678","citing_title":"Grounded Reinforcement Learning for Visual Reasoning","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20448","citing_title":"Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2507.01955","citing_title":"How Well Does GPT-4o Understand Vision? Evaluating Multimodal Foundation Models on Standard Computer Vision Tasks","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2511.16518","citing_title":"MiMo-Embodied: X-Embodied Foundation Model Technical Report","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08816","citing_title":"Mirror, Mirror on the Wall: Can VLM Agents Tell Who They Are at All?","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21921","citing_title":"Context Unrolling in Omni Models","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12357","citing_title":"ReflectCAP: Detailed Image Captioning with Reflective Memory","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2505.07062","citing_title":"Seed1.5-VL Technical Report","ref_index":109,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18512","citing_title":"S2H-DPO: Hardness-Aware Preference Optimization for Vision-Language Models","ref_index":118,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC","json":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC.json","graph_json":"https://pith.science/api/pith-number/BG6KOY6CM3RKLSDY6X2RSNMIAC/graph.json","events_json":"https://pith.science/api/pith-number/BG6KOY6CM3RKLSDY6X2RSNMIAC/events.json","paper":"https://pith.science/paper/BG6KOY6C"},"agent_actions":{"view_html":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC","download_json":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC.json","view_paper":"https://pith.science/paper/BG6KOY6C","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.06581&json=true","fetch_graph":"https://pith.science/api/pith-number/BG6KOY6CM3RKLSDY6X2RSNMIAC/graph.json","fetch_events":"https://pith.science/api/pith-number/BG6KOY6CM3RKLSDY6X2RSNMIAC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC/action/storage_attestation","attest_author":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC/action/author_attestation","sign_citation":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC/action/citation_signature","submit_replication":"https://pith.science/pith/BG6KOY6CM3RKLSDY6X2RSNMIAC/action/replication_record"}},"created_at":"2026-07-05T10:40:01.797186+00:00","updated_at":"2026-07-05T10:40:01.797186+00:00"}