{"paper":{"title":"Weather-Robust Cross-View Geo-Localization via Prototype-Based Semantic Part Discovery","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SkyPart discovers semantic parts in drone and satellite images using competing learnable prototypes to match views despite weather and altitude changes.","cross_cats":["cs.AI","cs.RO"],"primary_cat":"cs.CV","authors_text":"Chi-Nguyen Tran, Dao Sy Duy Minh, Huynh Trung Kiet, Long Tran-Thanh, Nguyen Lam Phu Quy, Phu-Hoa Pham","submitted_at":"2026-05-12T07:15:52Z","abstract_excerpt":"Cross-view geo-localization (CVGL), which matches an oblique drone view to a geo-referenced satellite tile, has emerged as a key alternative for autonomous drone navigation when GNSS signals are jammed, spoofed, or unavailable. Despite strong recent progress, three limitations persist: (1) global-descriptor designs compress the patch grid into a single vector without separating layout from texture across the view gap; (2) altitude-related scale variation is retained in the learned embedding rather than marginalized; and (3) multi-objective training relies on hand-tuned scalars over losses on i"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"At 26.95M parameters and 22.14 GFLOPs, SkyPart is the smallest among top-performing methods and sets a new state of the art on SUES-200, University-1652, and DenseUAV under a single-pass, no-re-ranking, no-TTA protocol. Its advantage over the strongest baseline widens under the ten-condition WeatherPrompt corruption benchmark.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That single-pass cosine assignment of patches to learnable prototypes will reliably discover semantic parts that separate layout from texture across the drastic view gap, and that altitude-conditioned modulation applied only during training will produce an altitude-invariant embedding at inference without loss of discriminative power.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SkyPart uses learnable prototypes for patch grouping, altitude modulation only in training, graph-attention readout, and Kendall-weighted loss to set new state-of-the-art single-pass performance on SUES-200, University-1652, and DenseUAV while widening gains under weather corruptions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SkyPart discovers semantic parts in drone and satellite images using competing learnable prototypes to match views despite weather and altitude changes.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2bdefa7e0f52a3372c5f82bbae5a0f97307c3fa0ac59d81479b5506a211eaa89"},"source":{"id":"2605.11654","kind":"arxiv","version":2},"verdict":{"id":"19b0181f-ca6b-4e6c-bf2d-77e17b427b1d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T01:26:12.072935Z","strongest_claim":"At 26.95M parameters and 22.14 GFLOPs, SkyPart is the smallest among top-performing methods and sets a new state of the art on SUES-200, University-1652, and DenseUAV under a single-pass, no-re-ranking, no-TTA protocol. Its advantage over the strongest baseline widens under the ten-condition WeatherPrompt corruption benchmark.","one_line_summary":"SkyPart uses learnable prototypes for patch grouping, altitude modulation only in training, graph-attention readout, and Kendall-weighted loss to set new state-of-the-art single-pass performance on SUES-200, University-1652, and DenseUAV while widening gains under weather corruptions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That single-pass cosine assignment of patches to learnable prototypes will reliably discover semantic parts that separate layout from texture across the drastic view gap, and that altitude-conditioned modulation applied only during training will produce an altitude-invariant embedding at inference without loss of discriminative power.","pith_extraction_headline":"SkyPart discovers semantic parts in drone and satellite images using competing learnable prototypes to match views despite weather and altitude changes."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11654/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T11:40:26.900415Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T09:31:17.707178Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:17:03.643559Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"ff6fe46f5296dd831de60139da67983cfb4359eee27e21d2772a29cdf5b35a1a"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"692b82c9bc6f581afb8b9094da1ece96eea2cc5d7a4935b66e1a10a266c19ec5"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}