{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:4KXQSEKCACNIJPCAS2CCXQUMFW","short_pith_number":"pith:4KXQSEKC","canonical_record":{"source":{"id":"2605.13667","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T15:27:41Z","cross_cats_sorted":[],"title_canon_sha256":"f04ec2be006a9833106289e9936921d6238aa3883a63199fe5bc8268919a8873","abstract_canon_sha256":"d07ffdb47a3c956d7a68e5aaa76b132c5c7815a7172b31fcc35853cc1d19a574"},"schema_version":"1.0"},"canonical_sha256":"e2af091142009a84bc4096842bc28c2db62a356946ac6d3cd57d974549e11f2f","source":{"kind":"arxiv","id":"2605.13667","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13667","created_at":"2026-05-18T02:44:17Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13667v1","created_at":"2026-05-18T02:44:17Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13667","created_at":"2026-05-18T02:44:17Z"},{"alias_kind":"pith_short_12","alias_value":"4KXQSEKCACNI","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"4KXQSEKCACNIJPCA","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"4KXQSEKC","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:4KXQSEKCACNIJPCAS2CCXQUMFW","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13667","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T15:27:41Z","cross_cats_sorted":[],"title_canon_sha256":"f04ec2be006a9833106289e9936921d6238aa3883a63199fe5bc8268919a8873","abstract_canon_sha256":"d07ffdb47a3c956d7a68e5aaa76b132c5c7815a7172b31fcc35853cc1d19a574"},"schema_version":"1.0"},"canonical_sha256":"e2af091142009a84bc4096842bc28c2db62a356946ac6d3cd57d974549e11f2f","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:17.232626Z","signature_b64":"Mwt/Iottin8e3OyE3+V6hqeQjTr8OsOAV5RDpBZFi7/uT3pDp78L/3wD9Cfs4CgVbuAGR74eO+rJyr9u/uj1Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e2af091142009a84bc4096842bc28c2db62a356946ac6d3cd57d974549e11f2f","last_reissued_at":"2026-05-18T02:44:17.232036Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:17.232036Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13667","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:17Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"62URuKd15anbwIdrXJqHZCuYvoKDS7mxyDm8/7mE4F5xrzxOXjy1zEBHzucTtyQctvWyIz9AdWpM+PmX0Hr+CA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-20T06:21:52.408779Z"},"content_sha256":"a8ace781542ad6a1b823f861180dc2870ba7656f8c1a395122880c5b94f3a50a","schema_version":"1.0","event_id":"sha256:a8ace781542ad6a1b823f861180dc2870ba7656f8c1a395122880c5b94f3a50a"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:4KXQSEKCACNIJPCAS2CCXQUMFW","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"SceneGraphVLM: Dynamic Scene Graph Generation from Video with Vision-Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SceneGraphVLM generates complete scene graphs from videos in about one second using compact vision-language models and targeted reinforcement learning.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Dmitry Yudin, Mark Gizetdinov, Vladislav Makarov","submitted_at":"2026-05-13T15:27:41Z","abstract_excerpt":"Scene graph generation provides a compact structured representation for visual perception, but accurate and fast graph prediction from images and videos remains challenging. Recent VLM-based methods can generate scene graphs end-to-end as structured text, yet often produce long outputs with irrelevant objects and relations. We present SceneGraphVLM, a compact method for image and video scene graph generation with small visual language models. SceneGraphVLM serializes graphs in a token-efficient TOON format and trains the model in two stages: supervised fine-tuning followed by reinforcement lea"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"With compact VLMs and vLLM-accelerated decoding, SceneGraphVLM achieves a strong quality-speed trade-off, improves precision-oriented SGG metrics while preserving reasonable recall, and generates complete scene graphs with approximately one-second latency.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the hallucination-aware RL rewards successfully balance coverage and precision on the target benchmarks without introducing new failure modes or requiring dataset-specific tuning that does not generalize.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SceneGraphVLM generates dynamic scene graphs from video using compact VLMs, TOON serialization, and hallucination-aware RL to improve precision and achieve one-second latency.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SceneGraphVLM generates complete scene graphs from videos in about one second using compact vision-language models and targeted reinforcement learning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"cdc73857edbf736eec9311abebb33e553a65960994222df0d88be1ee398d2d86"},"source":{"id":"2605.13667","kind":"arxiv","version":1},"verdict":{"id":"2ca0be25-326a-43c7-8530-f666df097943","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:36:53.643217Z","strongest_claim":"With compact VLMs and vLLM-accelerated decoding, SceneGraphVLM achieves a strong quality-speed trade-off, improves precision-oriented SGG metrics while preserving reasonable recall, and generates complete scene graphs with approximately one-second latency.","one_line_summary":"SceneGraphVLM generates dynamic scene graphs from video using compact VLMs, TOON serialization, and hallucination-aware RL to improve precision and achieve one-second latency.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the hallucination-aware RL rewards successfully balance coverage and precision on the target benchmarks without introducing new failure modes or requiring dataset-specific tuning that does not generalize.","pith_extraction_headline":"SceneGraphVLM generates complete scene graphs from videos in about one second using compact vision-language models and targeted reinforcement learning."},"references":{"count":48,"sample":[{"doi":"","year":2025,"title":"Qwen3-VL Technical Report","work_id":"1fe243aa-e3c0-4da6-b391-4cbcfc88d5c0","ref_index":1,"cited_arxiv_id":"2511.21631","is_internal_anchor":true},{"doi":"","year":null,"title":"Qwen2.5-vl technical report","work_id":"8cce8101-1a0a-4da5-a8d4-8b72b9d0c60f","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Qwen2.5-VL Technical Report","work_id":"69dffacb-bfe8-442d-be86-48624c60426f","ref_index":3,"cited_arxiv_id":"2502.13923","is_internal_anchor":true},{"doi":"","year":2024,"title":"Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling","work_id":"ee70bdc8-4656-4849-ada7-ce42a2278d70","ref_index":4,"cited_arxiv_id":"2412.05271","is_internal_anchor":true},{"doi":"","year":2025,"title":"Compile scene graphs with reinforcement learning","work_id":"bb1a8a98-c413-4df6-8e94-70284c64162f","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":48,"snapshot_sha256":"97a2867af96301a2557f855dc219193bd504a966b68631f2e0dc63e3ebe7d690","internal_anchors":11},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6a5a90d487e236227f8f97570d3224441a144ba0a757d583bc544af44218eb26"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"2ca0be25-326a-43c7-8530-f666df097943"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:17Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JhKAch1b09p2d/jVTKnIx5fnKom5x9Y6Upeil5BP3AqGyTQ15BCk5+IvvkKlLtyNHs9GCqIbLPJKrnViTJz8CQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-20T06:21:52.409329Z"},"content_sha256":"1815e6a2fd0450964335dfa1398c2ea71807b32c192781ed69b2cfbac528e057","schema_version":"1.0","event_id":"sha256:1815e6a2fd0450964335dfa1398c2ea71807b32c192781ed69b2cfbac528e057"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/4KXQSEKCACNIJPCAS2CCXQUMFW/bundle.json","state_url":"https://pith.science/pith/4KXQSEKCACNIJPCAS2CCXQUMFW/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/4KXQSEKCACNIJPCAS2CCXQUMFW/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-20T06:21:52Z","links":{"resolver":"https://pith.science/pith/4KXQSEKCACNIJPCAS2CCXQUMFW","bundle":"https://pith.science/pith/4KXQSEKCACNIJPCAS2CCXQUMFW/bundle.json","state":"https://pith.science/pith/4KXQSEKCACNIJPCAS2CCXQUMFW/state.json","well_known_bundle":"https://pith.science/.well-known/pith/4KXQSEKCACNIJPCAS2CCXQUMFW/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:4KXQSEKCACNIJPCAS2CCXQUMFW","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"d07ffdb47a3c956d7a68e5aaa76b132c5c7815a7172b31fcc35853cc1d19a574","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T15:27:41Z","title_canon_sha256":"f04ec2be006a9833106289e9936921d6238aa3883a63199fe5bc8268919a8873"},"schema_version":"1.0","source":{"id":"2605.13667","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13667","created_at":"2026-05-18T02:44:17Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13667v1","created_at":"2026-05-18T02:44:17Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13667","created_at":"2026-05-18T02:44:17Z"},{"alias_kind":"pith_short_12","alias_value":"4KXQSEKCACNI","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"4KXQSEKCACNIJPCA","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"4KXQSEKC","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:1815e6a2fd0450964335dfa1398c2ea71807b32c192781ed69b2cfbac528e057","target":"graph","created_at":"2026-05-18T02:44:17Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"With compact VLMs and vLLM-accelerated decoding, SceneGraphVLM achieves a strong quality-speed trade-off, improves precision-oriented SGG metrics while preserving reasonable recall, and generates complete scene graphs with approximately one-second latency."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the hallucination-aware RL rewards successfully balance coverage and precision on the target benchmarks without introducing new failure modes or requiring dataset-specific tuning that does not generalize."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SceneGraphVLM generates dynamic scene graphs from video using compact VLMs, TOON serialization, and hallucination-aware RL to improve precision and achieve one-second latency."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"SceneGraphVLM generates complete scene graphs from videos in about one second using compact vision-language models and targeted reinforcement learning."}],"snapshot_sha256":"cdc73857edbf736eec9311abebb33e553a65960994222df0d88be1ee398d2d86"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6a5a90d487e236227f8f97570d3224441a144ba0a757d583bc544af44218eb26"},"paper":{"abstract_excerpt":"Scene graph generation provides a compact structured representation for visual perception, but accurate and fast graph prediction from images and videos remains challenging. Recent VLM-based methods can generate scene graphs end-to-end as structured text, yet often produce long outputs with irrelevant objects and relations. We present SceneGraphVLM, a compact method for image and video scene graph generation with small visual language models. SceneGraphVLM serializes graphs in a token-efficient TOON format and trains the model in two stages: supervised fine-tuning followed by reinforcement lea","authors_text":"Dmitry Yudin, Mark Gizetdinov, Vladislav Makarov","cross_cats":[],"headline":"SceneGraphVLM generates complete scene graphs from videos in about one second using compact vision-language models and targeted reinforcement learning.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T15:27:41Z","title":"SceneGraphVLM: Dynamic Scene Graph Generation from Video with Vision-Language Models"},"references":{"count":48,"internal_anchors":11,"resolved_work":48,"sample":[{"cited_arxiv_id":"2511.21631","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Qwen3-VL Technical Report","work_id":"1fe243aa-e3c0-4da6-b391-4cbcfc88d5c0","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Qwen2.5-vl technical report","work_id":"8cce8101-1a0a-4da5-a8d4-8b72b9d0c60f","year":null},{"cited_arxiv_id":"2502.13923","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Qwen2.5-VL Technical Report","work_id":"69dffacb-bfe8-442d-be86-48624c60426f","year":null},{"cited_arxiv_id":"2412.05271","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling","work_id":"ee70bdc8-4656-4849-ada7-ce42a2278d70","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Compile scene graphs with reinforcement learning","work_id":"bb1a8a98-c413-4df6-8e94-70284c64162f","year":2025}],"snapshot_sha256":"97a2867af96301a2557f855dc219193bd504a966b68631f2e0dc63e3ebe7d690"},"source":{"id":"2605.13667","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:36:53.643217Z","id":"2ca0be25-326a-43c7-8530-f666df097943","model_set":{"reader":"grok-4.3"},"one_line_summary":"SceneGraphVLM generates dynamic scene graphs from video using compact VLMs, TOON serialization, and hallucination-aware RL to improve precision and achieve one-second latency.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"SceneGraphVLM generates complete scene graphs from videos in about one second using compact vision-language models and targeted reinforcement learning.","strongest_claim":"With compact VLMs and vLLM-accelerated decoding, SceneGraphVLM achieves a strong quality-speed trade-off, improves precision-oriented SGG metrics while preserving reasonable recall, and generates complete scene graphs with approximately one-second latency.","weakest_assumption":"That the hallucination-aware RL rewards successfully balance coverage and precision on the target benchmarks without introducing new failure modes or requiring dataset-specific tuning that does not generalize."}},"verdict_id":"2ca0be25-326a-43c7-8530-f666df097943"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:a8ace781542ad6a1b823f861180dc2870ba7656f8c1a395122880c5b94f3a50a","target":"record","created_at":"2026-05-18T02:44:17Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"d07ffdb47a3c956d7a68e5aaa76b132c5c7815a7172b31fcc35853cc1d19a574","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T15:27:41Z","title_canon_sha256":"f04ec2be006a9833106289e9936921d6238aa3883a63199fe5bc8268919a8873"},"schema_version":"1.0","source":{"id":"2605.13667","kind":"arxiv","version":1}},"canonical_sha256":"e2af091142009a84bc4096842bc28c2db62a356946ac6d3cd57d974549e11f2f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"e2af091142009a84bc4096842bc28c2db62a356946ac6d3cd57d974549e11f2f","first_computed_at":"2026-05-18T02:44:17.232036Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:44:17.232036Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Mwt/Iottin8e3OyE3+V6hqeQjTr8OsOAV5RDpBZFi7/uT3pDp78L/3wD9Cfs4CgVbuAGR74eO+rJyr9u/uj1Cg==","signature_status":"signed_v1","signed_at":"2026-05-18T02:44:17.232626Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13667","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:a8ace781542ad6a1b823f861180dc2870ba7656f8c1a395122880c5b94f3a50a","sha256:1815e6a2fd0450964335dfa1398c2ea71807b32c192781ed69b2cfbac528e057"],"state_sha256":"94ee63b15e9be3b62a5dc1188d8e9c8aa92d1f6cb3b8bbfada8d701b82d35440"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"IuwvJgJzGEbjrCBFBZEf4RhK02mk473UJPnD+zkcCLYhWGHZXkyK1Dg/65AQpn07ffth/RjziLk5cV6qZROkBA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-20T06:21:52.411682Z","bundle_sha256":"7cf1db2a92abf683689bf2f010f8899b16c0a9e922caf65525d7174fca59255b"}}