{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:UUNRMKTQJBN2ML3OIWS4AJ4UKG","short_pith_number":"pith:UUNRMKTQ","schema_version":"1.0","canonical_sha256":"a51b162a70485ba62f6e45a5c0279451a6010fa7d5c0953f72c97ec5c0a07065","source":{"kind":"arxiv","id":"2601.05729","version":2},"attestation_state":"computed","paper":{"title":"TAGRPO: Boosting GRPO on Image-to-Video Generation with Direct Trajectory Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Comi Chen, Guangzheng Xu, Haoyu Yang, Jianxiang Lu, Jin Wang, Linqing Wang, Longhuang Wu, Mingtao Chen, Peng Chen, Ping Luo, Qinglin Lu, Shuai Shao, Zhichao Hu","submitted_at":"2026-01-09T11:15:27Z","abstract_excerpt":"Recent studies have demonstrated the efficacy of integrating Group Relative Policy Optimization (GRPO) into flow matching models, particularly for text-to-image and text-to-video generation. However, we find that directly applying these techniques to image-to-video (I2V) models often fails to yield consistent reward improvements. To address this limitation, we present TAGRPO, a robust post-training framework for I2V models inspired by contrastive learning. Our approach is grounded in the observation that rollout videos generated from identical initial noise provide superior guidance for optimi"},"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":"2601.05729","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-01-09T11:15:27Z","cross_cats_sorted":[],"title_canon_sha256":"bc399b77017e7852be7e854337f48631deac9613ea76a4fb428e96f70db770c5","abstract_canon_sha256":"0ddb59d746a97ae0ba1ddf58b659cfd58fe783a73deb9c579cd9b710fadc4e14"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:05:42.290529Z","signature_b64":"VFyYT2GtTzaYuToY5iEB8DVutnF/alMe0OjddH/nGIv/Ktns8jtKtEhsMqZ0W5lar9aEt6c/wYW5I72y38JBBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a51b162a70485ba62f6e45a5c0279451a6010fa7d5c0953f72c97ec5c0a07065","last_reissued_at":"2026-05-27T01:05:42.289904Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:05:42.289904Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TAGRPO: Boosting GRPO on Image-to-Video Generation with Direct Trajectory Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Comi Chen, Guangzheng Xu, Haoyu Yang, Jianxiang Lu, Jin Wang, Linqing Wang, Longhuang Wu, Mingtao Chen, Peng Chen, Ping Luo, Qinglin Lu, Shuai Shao, Zhichao Hu","submitted_at":"2026-01-09T11:15:27Z","abstract_excerpt":"Recent studies have demonstrated the efficacy of integrating Group Relative Policy Optimization (GRPO) into flow matching models, particularly for text-to-image and text-to-video generation. However, we find that directly applying these techniques to image-to-video (I2V) models often fails to yield consistent reward improvements. To address this limitation, we present TAGRPO, a robust post-training framework for I2V models inspired by contrastive learning. Our approach is grounded in the observation that rollout videos generated from identical initial noise provide superior guidance for optimi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.05729","kind":"arxiv","version":2},"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/2601.05729/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":"2601.05729","created_at":"2026-05-27T01:05:42.289977+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.05729v2","created_at":"2026-05-27T01:05:42.289977+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.05729","created_at":"2026-05-27T01:05:42.289977+00:00"},{"alias_kind":"pith_short_12","alias_value":"UUNRMKTQJBN2","created_at":"2026-05-27T01:05:42.289977+00:00"},{"alias_kind":"pith_short_16","alias_value":"UUNRMKTQJBN2ML3O","created_at":"2026-05-27T01:05:42.289977+00:00"},{"alias_kind":"pith_short_8","alias_value":"UUNRMKTQ","created_at":"2026-05-27T01:05:42.289977+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2605.17248","citing_title":"Image-to-Video Diffusion: From Foundations to Open Frontiers","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.14274","citing_title":"CreFlow: Corrective Reflow for Sparse-Reward Embodied Video Diffusion RL","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11723","citing_title":"CaC: Advancing Video Reward Models via Hierarchical Spatiotemporal Concentrating","ref_index":44,"is_internal_anchor":true},{"citing_arxiv_id":"2604.14910","citing_title":"Reward-Aware Trajectory Shaping for Few-step Visual Generation","ref_index":40,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG","json":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG.json","graph_json":"https://pith.science/api/pith-number/UUNRMKTQJBN2ML3OIWS4AJ4UKG/graph.json","events_json":"https://pith.science/api/pith-number/UUNRMKTQJBN2ML3OIWS4AJ4UKG/events.json","paper":"https://pith.science/paper/UUNRMKTQ"},"agent_actions":{"view_html":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG","download_json":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG.json","view_paper":"https://pith.science/paper/UUNRMKTQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.05729&json=true","fetch_graph":"https://pith.science/api/pith-number/UUNRMKTQJBN2ML3OIWS4AJ4UKG/graph.json","fetch_events":"https://pith.science/api/pith-number/UUNRMKTQJBN2ML3OIWS4AJ4UKG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG/action/storage_attestation","attest_author":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG/action/author_attestation","sign_citation":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG/action/citation_signature","submit_replication":"https://pith.science/pith/UUNRMKTQJBN2ML3OIWS4AJ4UKG/action/replication_record"}},"created_at":"2026-05-27T01:05:42.289977+00:00","updated_at":"2026-05-27T01:05:42.289977+00:00"}