{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:EVIGEFXKTZQOFFQLZ6Y5XWO36Z","short_pith_number":"pith:EVIGEFXK","schema_version":"1.0","canonical_sha256":"25506216ea9e60e2960bcfb1dbd9dbf6757bdab43fad1292f5e4c1cdde2ad65c","source":{"kind":"arxiv","id":"2605.13803","version":1},"attestation_state":"computed","paper":{"title":"EvoGround: Self-Evolving Video Agents for Video Temporal Grounding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Two self-evolving agents learn video temporal grounding from unlabeled videos alone.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Byoung-Tak Zhang, Lorenzo Torresani, Minjoon Jung","submitted_at":"2026-05-13T17:25:51Z","abstract_excerpt":"Video temporal grounding (VTG) takes an untrimmed video and a natural-language query as input and localizes the temporal moment that best matches the query. Existing methods rely on large, task-specific datasets requiring costly manual annotation. We introduce EvoGround, a framework of two coupled self-evolving agents, a proposer and a solver, that learn temporal grounding from raw videos without any human-labeled data. The proposer generates query--moment pairs from raw videos, while the solver learns to ground them and feeds back signals that improve the proposer in return. Through this self"},"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":"2605.13803","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T17:25:51Z","cross_cats_sorted":[],"title_canon_sha256":"1f3c58016865f2d0427d7bdf33660b4bf2d489be65239d9f31094989d1ebb763","abstract_canon_sha256":"114145067ac93f8f12e153b1a05c767d2e23e58d496978b752faf6d5978ee62d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:15.488428Z","signature_b64":"88sSFY06jPXwswz1FzBMQUOO+CwFqSZKBidzOsJz0TRgUkekswtiZMz3Mkgi6qhwfuCvLS5ES2N/DWCor8aABw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"25506216ea9e60e2960bcfb1dbd9dbf6757bdab43fad1292f5e4c1cdde2ad65c","last_reissued_at":"2026-05-18T02:44:15.487853Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:15.487853Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"EvoGround: Self-Evolving Video Agents for Video Temporal Grounding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Two self-evolving agents learn video temporal grounding from unlabeled videos alone.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Byoung-Tak Zhang, Lorenzo Torresani, Minjoon Jung","submitted_at":"2026-05-13T17:25:51Z","abstract_excerpt":"Video temporal grounding (VTG) takes an untrimmed video and a natural-language query as input and localizes the temporal moment that best matches the query. Existing methods rely on large, task-specific datasets requiring costly manual annotation. We introduce EvoGround, a framework of two coupled self-evolving agents, a proposer and a solver, that learn temporal grounding from raw videos without any human-labeled data. The proposer generates query--moment pairs from raw videos, while the solver learns to ground them and feeds back signals that improve the proposer in return. Through this self"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Trained on 2.5K unlabeled videos, EvoGround matches or surpasses fully supervised models across multiple VTG benchmarks, while emerging as a state-of-the-art fine-grained video captioner without manual labels.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The mutual reinforcement loop between proposer and solver can bootstrap effective temporal grounding and captioning capabilities starting from raw videos and a shared backbone without any initial human supervision or external reward signals.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A proposer-solver agent pair achieves supervised-level video temporal grounding and fine-grained captioning from 2.5K unlabeled videos via self-reinforcing evolution.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Two self-evolving agents learn video temporal grounding from unlabeled videos alone.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"796395d485ef06834c1c052454b595f7113f6af39759cddc7b27bb5ce0c840c0"},"source":{"id":"2605.13803","kind":"arxiv","version":1},"verdict":{"id":"b39179a1-f7bf-4004-b335-63c64be842b7","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:24:40.631477Z","strongest_claim":"Trained on 2.5K unlabeled videos, EvoGround matches or surpasses fully supervised models across multiple VTG benchmarks, while emerging as a state-of-the-art fine-grained video captioner without manual labels.","one_line_summary":"A proposer-solver agent pair achieves supervised-level video temporal grounding and fine-grained captioning from 2.5K unlabeled videos via self-reinforcing evolution.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The mutual reinforcement loop between proposer and solver can bootstrap effective temporal grounding and captioning capabilities starting from raw videos and a shared backbone without any initial human supervision or external reward signals.","pith_extraction_headline":"Two self-evolving agents learn video temporal grounding from unlabeled videos alone."},"references":{"count":69,"sample":[{"doi":"","year":2022,"title":"Modal-specific pseudo query generation for video corpus moment retrieval,","work_id":"8976c7d1-e459-4a52-a8f2-30bdf3ff6019","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Detecting moments and highlights in videos via natural language queries,","work_id":"28515618-525c-4594-9aa4-d398980a7891","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Can i trust your answer? visually grounded video question answering,","work_id":"534671d4-6ee8-42c5-baeb-15da71cc7c0b","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models","work_id":"7a50d850-6338-479d-ba02-65efc1b03f01","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis","work_id":"77fd5ac9-ae98-4846-9d83-e9c73c8f2a52","ref_index":5,"cited_arxiv_id":"2405.21075","is_internal_anchor":true}],"resolved_work":69,"snapshot_sha256":"beb31c4f201c0196644436bf899bd157d6c3b74f04f0bf44e8437ff822f38214","internal_anchors":20},"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":"2605.13803","created_at":"2026-05-18T02:44:15.487937+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.13803v1","created_at":"2026-05-18T02:44:15.487937+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13803","created_at":"2026-05-18T02:44:15.487937+00:00"},{"alias_kind":"pith_short_12","alias_value":"EVIGEFXKTZQO","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"EVIGEFXKTZQOFFQL","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"EVIGEFXK","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/EVIGEFXKTZQOFFQLZ6Y5XWO36Z","json":"https://pith.science/pith/EVIGEFXKTZQOFFQLZ6Y5XWO36Z.json","graph_json":"https://pith.science/api/pith-number/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/graph.json","events_json":"https://pith.science/api/pith-number/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/events.json","paper":"https://pith.science/paper/EVIGEFXK"},"agent_actions":{"view_html":"https://pith.science/pith/EVIGEFXKTZQOFFQLZ6Y5XWO36Z","download_json":"https://pith.science/pith/EVIGEFXKTZQOFFQLZ6Y5XWO36Z.json","view_paper":"https://pith.science/paper/EVIGEFXK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.13803&json=true","fetch_graph":"https://pith.science/api/pith-number/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/graph.json","fetch_events":"https://pith.science/api/pith-number/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/action/storage_attestation","attest_author":"https://pith.science/pith/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/action/author_attestation","sign_citation":"https://pith.science/pith/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/action/citation_signature","submit_replication":"https://pith.science/pith/EVIGEFXKTZQOFFQLZ6Y5XWO36Z/action/replication_record"}},"created_at":"2026-05-18T02:44:15.487937+00:00","updated_at":"2026-05-18T02:44:15.487937+00:00"}