{"paper":{"title":"ChangeFlow -- Latent Rectified Flow for Change Detection in Remote Sensing","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Remote sensing change detection improves by generating distributions of plausible masks in latent space with a rectified flow model.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Bla\\v{z} Rolih, Filip Wolf, Luka \\v{C}ehovin Zajc, Matic Fu\\v{c}ka","submitted_at":"2026-05-14T20:04:16Z","abstract_excerpt":"Remote sensing change detection (RSCD) aims to localise changes between two images of the same geographic region. In practice, change masks often follow region-level annotation conventions rather than purely local appearance differences, making them context-dependent and occasionally ambiguous. Most state-of-the-art methods utilise per-pixel discriminative classification, which produces a single prediction per input and fails to explicitly model the changed region as a coherent whole. A natural alternative is generative formulation, which can model a distribution of plausible masks, enabling s"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across four benchmarks, ChangeFlow achieves an average F1 of 80.4%, improving by 1.3 points on average over the previous best method, while maintaining inference speed comparable to recent strong baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a structured yet lightweight conditioning signal in latent space is sufficient for the rectified-flow model to capture both global consistency of changed regions and the distribution of plausible masks that reflect annotation ambiguity.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ChangeFlow reformulates remote sensing change detection as latent rectified-flow mask synthesis, reaching 80.4% average F1 across four benchmarks with 1.3-point gain and sampling-based ensembling.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Remote sensing change detection improves by generating distributions of plausible masks in latent space with a rectified flow model.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9993fa295ea4a20bae8b219a5988354d0dcf04ca240d07803adf10f56c87bec2"},"source":{"id":"2605.15375","kind":"arxiv","version":1},"verdict":{"id":"856c2f9e-c0cb-4991-a940-4f778b2caabf","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T15:47:57.955221Z","strongest_claim":"Across four benchmarks, ChangeFlow achieves an average F1 of 80.4%, improving by 1.3 points on average over the previous best method, while maintaining inference speed comparable to recent strong baselines.","one_line_summary":"ChangeFlow reformulates remote sensing change detection as latent rectified-flow mask synthesis, reaching 80.4% average F1 across four benchmarks with 1.3-point gain and sampling-based ensembling.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a structured yet lightweight conditioning signal in latent space is sufficient for the rectified-flow model to capture both global consistency of changed regions and the distribution of plausible masks that reflect annotation ambiguity.","pith_extraction_headline":"Remote sensing change detection improves by generating distributions of plausible masks in latent space with a rectified flow model."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15375/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T16:01:18.055328Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:53:41.691717Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T14:21:54.182428Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.733083Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"2e728b96eeb19ef706fe9f6535f55b00b9160e9c10624338dec84e78ba602de4"},"references":{"count":62,"sample":[{"doi":"","year":2016,"title":"Layer Normalization","work_id":"20a2d720-0046-4c7c-bcd6-327ec8143f69","ref_index":1,"cited_arxiv_id":"1607.06450","is_internal_anchor":true},{"doi":"","year":2025,"title":"In: Proceedings of the IEEE/CVF International Conference on Computer Vision","work_id":"5bb70ef9-71e0-45c6-8fe0-8969ac78007c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"In: IEEE/CVF Winter Conference on Applications of Computer Vision","work_id":"0b8232e4-84f4-4db2-a7f9-6d86165aedfb","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"In: IEEE International Geoscience and Remote Sensing Symposium","work_id":"9d4d5676-5462-46c6-81ac-599364fd8ac8","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"arXiv preprint arXiv:2512.05140 (2025) 4","work_id":"75cbc4ce-73e8-44e5-af33-de040081f535","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":62,"snapshot_sha256":"a0ae4bb4d43c9b2f4da8d5747663c47026f910d694d998a7d7cb8109799e8ba8","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"cb7a7311bead9a0ddf74eff7a93781719e8db64c00bd8b13c4c742a15345130a"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}