{"paper":{"title":"SynFlow: Scaling Up LiDAR Scene Flow Estimation with Synthetic Data","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Models trained only on synthetic LiDAR scene flow data match or beat real supervised baselines on multiple benchmarks.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Chenhan Jiang, Patric Jensfelt, Qingwen Zhang, Xiaomeng Zhu","submitted_at":"2026-04-10T15:25:33Z","abstract_excerpt":"Reliable 3D dynamic perception requires models that can anticipate motion beyond predefined categories, yet progress is hindered by the scarcity of dense, high-quality motion annotations. While self-supervision on unlabeled real data offers a path forward, empirical evidence suggests that scaling unlabeled data fails to close the performance gap due to noisy proxy signals. In this paper, we propose learning robust real-world motion priors entirely from scalable simulation. We introduce SynFlow, a data generation pipeline for large-scale synthetic LiDAR scene flow. Unlike prior works that prior"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Models trained exclusively on SynFlow-4k generalize across multiple real-world benchmarks in a zero-shot regime, rivaling in-domain supervised baselines on nuScenes and outperforming state-of-the-art methods on TruckScenes by 31.8%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The synthetic motion patterns generated by the pipeline are sufficiently representative of real-world kinematic distributions that models can learn domain-invariant priors without explicit domain adaptation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SynFlow creates a 34-times larger synthetic LiDAR scene flow dataset that lets models trained only on simulation match or beat supervised real-data baselines on multiple benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Models trained only on synthetic LiDAR scene flow data match or beat real supervised baselines on multiple benchmarks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"67f048427326bc8d26176c07c9b29485f617148dcb622e080a68f20edba66541"},"source":{"id":"2604.09411","kind":"arxiv","version":2},"verdict":{"id":"89315fb4-017f-41e9-b28f-9a91310a29e5","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T16:41:12.457339Z","strongest_claim":"Models trained exclusively on SynFlow-4k generalize across multiple real-world benchmarks in a zero-shot regime, rivaling in-domain supervised baselines on nuScenes and outperforming state-of-the-art methods on TruckScenes by 31.8%.","one_line_summary":"SynFlow creates a 34-times larger synthetic LiDAR scene flow dataset that lets models trained only on simulation match or beat supervised real-data baselines on multiple benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The synthetic motion patterns generated by the pipeline are sufficiently representative of real-world kinematic distributions that models can learn domain-invariant priors without explicit domain adaptation.","pith_extraction_headline":"Models trained only on synthetic LiDAR scene flow data match or beat real supervised baselines on multiple benchmarks."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.09411/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"}