{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GE2DIR2QLUPWAXS56AQ3PMYF6M","short_pith_number":"pith:GE2DIR2Q","schema_version":"1.0","canonical_sha256":"31343447505d1f605e5df021b7b305f3025e339a2b83095c51f7828270feb58b","source":{"kind":"arxiv","id":"2605.14709","version":1},"attestation_state":"computed","paper":{"title":"Breaking Dual Bottlenecks: Evolving Unified Multimodal Models into Self-Adaptive Interleaved Visual Reasoners","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Unified multimodal models learn to switch autonomously between direct generation, reflection, and planning to close the understanding-generation gap in image tasks.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bingjie Gao, Canmiao Fu, Chen Li, Feng Wang, Jiangtong Li, Keming Ye, Li Niu, Qingyang Liu, Shaobo Wang, Shuochen Chang, Yali Wang, Zhipeng Huang","submitted_at":"2026-05-14T11:27:46Z","abstract_excerpt":"Recent unified models integrate multimodal understanding and generation within a single framework. However, an \"understanding-generation gap\" persists, where models can capture user intent but often fail to translate this semantic knowledge into precise pixel-level manipulation. This gap results in two bottlenecks in anything-to-image task (X2I): the attention entanglement bottleneck, where blind planning struggles with complex prompts, and the visual refinement bottleneck, where unstructured feedback fails to correct imperfections efficiently. In this paper, we propose a novel framework that "},"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":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.14709","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-14T11:27:46Z","cross_cats_sorted":[],"title_canon_sha256":"3f9da7b42bbb9de9deeb4fe6777ad1d8822702621470cbc499065b556526684c","abstract_canon_sha256":"d56fe6ecb3a8caf23fc7cfd2ce12e7db1133bbd87c3d8c74c283c9ad2a5c5af5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:59.234645Z","signature_b64":"wBAKn/sYrbfR/qz38HUkvAPN6gm6stZEKQGKn1db6B3dEjvaecz5neGMnQ+JdlBpD5AT5OGDe7mIoCNFMp3dBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"31343447505d1f605e5df021b7b305f3025e339a2b83095c51f7828270feb58b","last_reissued_at":"2026-05-17T23:38:59.234134Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:59.234134Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Breaking Dual Bottlenecks: Evolving Unified Multimodal Models into Self-Adaptive Interleaved Visual Reasoners","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Unified multimodal models learn to switch autonomously between direct generation, reflection, and planning to close the understanding-generation gap in image tasks.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bingjie Gao, Canmiao Fu, Chen Li, Feng Wang, Jiangtong Li, Keming Ye, Li Niu, Qingyang Liu, Shaobo Wang, Shuochen Chang, Yali Wang, Zhipeng Huang","submitted_at":"2026-05-14T11:27:46Z","abstract_excerpt":"Recent unified models integrate multimodal understanding and generation within a single framework. However, an \"understanding-generation gap\" persists, where models can capture user intent but often fail to translate this semantic knowledge into precise pixel-level manipulation. This gap results in two bottlenecks in anything-to-image task (X2I): the attention entanglement bottleneck, where blind planning struggles with complex prompts, and the visual refinement bottleneck, where unstructured feedback fails to correct imperfections efficiently. In this paper, we propose a novel framework that "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"our method outperforms existing baselines on X2I, achieving superior generation fidelity among simple-to-complex instructions.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The constructed hierarchical data pipeline and designed step-wise rewards plus complexity penalty will enable effective autonomous mode switching without creating new bottlenecks or overfitting to the new dataset.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Unified multimodal models gain self-adaptive modes (direct generation, self-reflection, multi-step planning) trained via SFT and RL with step-wise rewards to close the understanding-generation gap in anything-to-image tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Unified multimodal models learn to switch autonomously between direct generation, reflection, and planning to close the understanding-generation gap in image tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"19f2cdc0898cd1d8245cd4eeec758901dc0d49d86dfd748b55c96957796192d0"},"source":{"id":"2605.14709","kind":"arxiv","version":1},"verdict":{"id":"ceb3a763-2059-41cb-9dd6-1a80e63d63eb","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T04:47:27.286732Z","strongest_claim":"our method outperforms existing baselines on X2I, achieving superior generation fidelity among simple-to-complex instructions.","one_line_summary":"Unified multimodal models gain self-adaptive modes (direct generation, self-reflection, multi-step planning) trained via SFT and RL with step-wise rewards to close the understanding-generation gap in anything-to-image tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The constructed hierarchical data pipeline and designed step-wise rewards plus complexity penalty will enable effective autonomous mode switching without creating new bottlenecks or overfitting to the new dataset.","pith_extraction_headline":"Unified multimodal models learn to switch autonomously between direct generation, reflection, and planning to close the understanding-generation gap in image tasks."},"references":{"count":36,"sample":[{"doi":"","year":2026,"title":"Thinking-while-generating: Interleaving tex- tual reasoning throughout visual generation.arXiv preprint arXiv:2511.16671, 2025","work_id":"4216b182-04a6-4cc1-a898-5aaf370461c3","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Your Objective: Evaluate how faithfully the Generated Image (Y) fulfills the **Instruction**, focusing on whether the requested changes or additions were executed correctly","work_id":"75a3f053-0bdd-4da5-8561-5cc7940f7220","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"**Detect Change**: What has been added, modified, or created in Y compared to X? (If X is Text-only, evaluate Y directly against the text)","work_id":"aaacd748-0e84-4cff-92a3-310083988a91","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"**Expected Visual Caption**: Describe the ideal result if the instruction were perfectly followed","work_id":"bff7bb31-bdbd-461e-9c21-4e0f996534bc","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"**Instruction Match**: - Was the correct subject/attribute modified or created? - For **Spatial/Size** changes: Is the placement or scale correct relative to the instruction? - For **Subject-driven** ","work_id":"93309ce4-e4b7-410c-8853-3d326e277ecb","ref_index":7,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":36,"snapshot_sha256":"dee9141e29d50e0c38e6a73ffaf497d33e34a1180894ed3d1411082df59ead52","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":"2605.14709","created_at":"2026-05-17T23:38:59.234215+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.14709v1","created_at":"2026-05-17T23:38:59.234215+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14709","created_at":"2026-05-17T23:38:59.234215+00:00"},{"alias_kind":"pith_short_12","alias_value":"GE2DIR2QLUPW","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"GE2DIR2QLUPWAXS5","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"GE2DIR2Q","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M","json":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M.json","graph_json":"https://pith.science/api/pith-number/GE2DIR2QLUPWAXS56AQ3PMYF6M/graph.json","events_json":"https://pith.science/api/pith-number/GE2DIR2QLUPWAXS56AQ3PMYF6M/events.json","paper":"https://pith.science/paper/GE2DIR2Q"},"agent_actions":{"view_html":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M","download_json":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M.json","view_paper":"https://pith.science/paper/GE2DIR2Q","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.14709&json=true","fetch_graph":"https://pith.science/api/pith-number/GE2DIR2QLUPWAXS56AQ3PMYF6M/graph.json","fetch_events":"https://pith.science/api/pith-number/GE2DIR2QLUPWAXS56AQ3PMYF6M/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M/action/storage_attestation","attest_author":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M/action/author_attestation","sign_citation":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M/action/citation_signature","submit_replication":"https://pith.science/pith/GE2DIR2QLUPWAXS56AQ3PMYF6M/action/replication_record"}},"created_at":"2026-05-17T23:38:59.234215+00:00","updated_at":"2026-05-17T23:38:59.234215+00:00"}