{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:JMFMACRYRWHZTMAAVBMP4GT44K","short_pith_number":"pith:JMFMACRY","schema_version":"1.0","canonical_sha256":"4b0ac00a388d8f99b000a858fe1a7ce28414943933cd7d2dba8323c97733eb1a","source":{"kind":"arxiv","id":"2405.02246","version":1},"attestation_state":"computed","paper":{"title":"What matters when building vision-language models?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Hugo Lauren\\c{c}on, L\\'eo Tronchon, Matthieu Cord, Victor Sanh","submitted_at":"2024-05-03T17:00:00Z","abstract_excerpt":"The growing interest in vision-language models (VLMs) has been driven by improvements in large language models and vision transformers. Despite the abundance of literature on this subject, we observe that critical decisions regarding the design of VLMs are often not justified. We argue that these unsupported decisions impede progress in the field by making it difficult to identify which choices improve model performance. To address this issue, we conduct extensive experiments around pre-trained models, architecture choice, data, and training methods. Our consolidation of findings includes the "},"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":"2405.02246","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2024-05-03T17:00:00Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"8536d4f5b642eecfa0533c9714f162a76f7d15640cfabf9504c059142b479d2a","abstract_canon_sha256":"875c63e21164f8796287b682c6a14d70baa67c2e0208a37378b83bcd75efdd18"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:15:08.818391Z","signature_b64":"oiSakkfhHpV1Ue2KC7Jf67LWrYqhDulI3BaljaahXYmtUB8YWTghch7xTcOOLmHEdNwY/o/xs7oyRPoMApZ2Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4b0ac00a388d8f99b000a858fe1a7ce28414943933cd7d2dba8323c97733eb1a","last_reissued_at":"2026-07-05T08:15:08.817931Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:15:08.817931Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"What matters when building vision-language models?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Hugo Lauren\\c{c}on, L\\'eo Tronchon, Matthieu Cord, Victor Sanh","submitted_at":"2024-05-03T17:00:00Z","abstract_excerpt":"The growing interest in vision-language models (VLMs) has been driven by improvements in large language models and vision transformers. Despite the abundance of literature on this subject, we observe that critical decisions regarding the design of VLMs are often not justified. We argue that these unsupported decisions impede progress in the field by making it difficult to identify which choices improve model performance. To address this issue, we conduct extensive experiments around pre-trained models, architecture choice, data, and training methods. Our consolidation of findings includes the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.02246","kind":"arxiv","version":1},"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/2405.02246/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":"2405.02246","created_at":"2026-07-05T08:15:08.817986+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.02246v1","created_at":"2026-07-05T08:15:08.817986+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.02246","created_at":"2026-07-05T08:15:08.817986+00:00"},{"alias_kind":"pith_short_12","alias_value":"JMFMACRYRWHZ","created_at":"2026-07-05T08:15:08.817986+00:00"},{"alias_kind":"pith_short_16","alias_value":"JMFMACRYRWHZTMAA","created_at":"2026-07-05T08:15:08.817986+00:00"},{"alias_kind":"pith_short_8","alias_value":"JMFMACRY","created_at":"2026-07-05T08:15:08.817986+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":25,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.02089","citing_title":"ESC: Emotional Self-Correction for Reliable Vision-Language Models","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11853","citing_title":"Task-Aware Structured Memory for Dynamic Multi-modal In-Context Learning","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09132","citing_title":"Vision Language Model Helps Private Information De-Identification in Vision Data","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09125","citing_title":"Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges","ref_index":103,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00606","citing_title":"Retrieved Images as Visual Thought: Training-Free Multimodal In-Context Learning for the Open-vs-Closed Gap","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28818","citing_title":"VLMs May Not Globally Enhance Human Alignment over LLMs During Natural Reading","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2408.04840","citing_title":"mPLUG-Owl3: Towards Long Image-Sequence Understanding in Multi-Modal Large Language Models","ref_index":218,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19866","citing_title":"Structured Layout Priors for Robust Out-of-Distribution Visual Document Understanding","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2412.14164","citing_title":"MetaMorph: Multimodal Understanding and Generation via Instruction Tuning","ref_index":267,"is_internal_anchor":false},{"citing_arxiv_id":"2406.16860","citing_title":"Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2410.10594","citing_title":"VisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documents","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2412.03555","citing_title":"PaliGemma 2: A Family of Versatile VLMs for Transfer","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2604.00799","citing_title":"Multimodal Language Models Cannot Spot Spatial Inconsistencies","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03231","citing_title":"CoME-VL: Scaling Complementary Multi-Encoder Vision-Language Learning","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2504.05299","citing_title":"SmolVLM: Redefining small and efficient multimodal models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21232","citing_title":"ReCAPA: Hierarchical Predictive Correction to Mitigate Cascading Failures","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2506.01844","citing_title":"SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06481","citing_title":"OA-WAM: Object-Addressable World Action Model for Robust Robot Manipulation","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2407.07726","citing_title":"PaliGemma: A versatile 3B VLM for transfer","ref_index":69,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11283","citing_title":"Multimodal Large Language Model-Enabled Video Translation: A Role-Oriented Survey","ref_index":182,"is_internal_anchor":false},{"citing_arxiv_id":"2407.07895","citing_title":"LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2408.01800","citing_title":"MiniCPM-V: A GPT-4V Level MLLM on Your Phone","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13803","citing_title":"Gaslight, Gatekeep, V1-V3: Early Visual Cortex Alignment Shields Vision-Language Models from Sycophantic Manipulation","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17949","citing_title":"ZSG-IAD: A Multimodal Framework for Zero-Shot Grounded Industrial Anomaly Detection","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21232","citing_title":"ReCAPA: Hierarchical Predictive Correction to Mitigate Cascading Failures","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K","json":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K.json","graph_json":"https://pith.science/api/pith-number/JMFMACRYRWHZTMAAVBMP4GT44K/graph.json","events_json":"https://pith.science/api/pith-number/JMFMACRYRWHZTMAAVBMP4GT44K/events.json","paper":"https://pith.science/paper/JMFMACRY"},"agent_actions":{"view_html":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K","download_json":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K.json","view_paper":"https://pith.science/paper/JMFMACRY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.02246&json=true","fetch_graph":"https://pith.science/api/pith-number/JMFMACRYRWHZTMAAVBMP4GT44K/graph.json","fetch_events":"https://pith.science/api/pith-number/JMFMACRYRWHZTMAAVBMP4GT44K/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K/action/storage_attestation","attest_author":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K/action/author_attestation","sign_citation":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K/action/citation_signature","submit_replication":"https://pith.science/pith/JMFMACRYRWHZTMAAVBMP4GT44K/action/replication_record"}},"created_at":"2026-07-05T08:15:08.817986+00:00","updated_at":"2026-07-05T08:15:08.817986+00:00"}