{"paper":{"title":"SHED: Style-Homogenized Embedding Alignment for Domain Generalization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Removing style centroids from CLIP embeddings improves generalization to unseen domains.","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Kai Gan, Tong Wei","submitted_at":"2026-05-16T12:51:38Z","abstract_excerpt":"Domain generalization aims to enhance model robustness against unseen domains with embedding distribution shifts. While large-scale vision-language models like CLIP exhibit strong generalization, their direct image-text embedding alignment suffers from inherent information asymmetry: images encode both class semantics and domain-specific styles, whereas text prompts primarily convey basic class cues. This asymmetry hinders generalization to novel domains in realistic scenarios. To address this, we propose Style-Homogenized Embedding alignment for Domain-generalization (SHED), a novel CLIP-base"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"SHED achieves state-of-the-art performance, outperforming prior methods significantly (e.g., +4.0% on DomainNet vs. standard fine-tuning).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That subtracting estimated style centroids from embeddings preserves class semantics while removing only domain-specific information, and that projecting textual domain centroids into visual space enables effective membership-weighted inference without any target domain data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SHED improves domain generalization in CLIP by aligning style-homogenized embeddings instead of raw ones, achieving state-of-the-art results on five benchmarks including a 4% gain on DomainNet.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Removing style centroids from CLIP embeddings improves generalization to unseen domains.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"97acd1be920168a93a565869c2cb48acccbf26a8206d2d170f150871707f6d06"},"source":{"id":"2605.16973","kind":"arxiv","version":1},"verdict":{"id":"38c0bc99-024c-421e-bf55-91dfe7df1b4d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:20:02.015623Z","strongest_claim":"SHED achieves state-of-the-art performance, outperforming prior methods significantly (e.g., +4.0% on DomainNet vs. standard fine-tuning).","one_line_summary":"SHED improves domain generalization in CLIP by aligning style-homogenized embeddings instead of raw ones, achieving state-of-the-art results on five benchmarks including a 4% gain on DomainNet.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That subtracting estimated style centroids from embeddings preserves class semantics while removing only domain-specific information, and that projecting textual domain centroids into visual space enables effective membership-weighted inference without any target domain data.","pith_extraction_headline":"Removing style centroids from CLIP embeddings improves generalization to unseen domains."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16973/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T20:31:33.579607Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T20:31:19.052947Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T19:51:56.906160Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"citation_quote_validity","ran_at":"2026-05-19T19:50:07.816280Z","status":"skipped","version":"0.1.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.221152Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.307823Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"ea1dea97c46bedc16014f49cd9520e47dae89572ea7bd80ba02670a33c2f349d"},"references":{"count":55,"sample":[{"doi":"","year":1985,"title":"Structure and Interpretation of Computer Programs","work_id":"01a14db6-7d1d-4607-868a-f7393285bd98","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2001,"title":"Visual Information Extraction with Lixto","work_id":"70a0bb1e-35f7-444f-b12d-bd6dcd37cc57","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1985,"title":"Brachman and James G","work_id":"0dc6cf5e-34e0-4f51-b711-0498f8e618ec","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1992,"title":"Complexity results for nonmonotonic logics","work_id":"d0459553-e996-400d-b3ca-0618c3da598c","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2002,"title":"Hypertree Decompositions and Tractable Queries","work_id":"97611438-476f-473e-b229-ee75292c4fac","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":55,"snapshot_sha256":"fc8fff5149a06699b8a2bf8c6361063ea54efc2a6190a74d2b0594b88c91d78b","internal_anchors":2},"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"}