{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NTVX2QFGGXF3YBDNID45X54RGW","short_pith_number":"pith:NTVX2QFG","schema_version":"1.0","canonical_sha256":"6ceb7d40a635cbbc046d40f9dbf791359730559d773de73f6faa62963527dde1","source":{"kind":"arxiv","id":"2411.15633","version":4},"attestation_state":"computed","paper":{"title":"Orthogonal Subspace Decomposition for Generalizable AI-Generated Image Detection","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Decomposing features via SVD into orthogonal parts lets detectors freeze general pre-trained knowledge and adapt only the rest to spot AI fakes without overfitting.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Baoyuan Wu, Chengchun Liu, Jiangming Wang, Ke-Yue Zhang, Li Yuan, Peng Jin, Shen Chen, Shouhong Ding, Taiping Yao, Zhiyuan Yan","submitted_at":"2024-11-23T19:10:32Z","abstract_excerpt":"AI-generated images (AIGIs), such as natural or face images, have become increasingly important yet challenging. In this paper, we start from a new perspective to excavate the reason behind the failure generalization in AIGI detection, named the \\textit{asymmetry phenomenon}, where a naively trained detector tends to favor overfitting to the limited and monotonous fake patterns, causing the feature space to become highly constrained and low-ranked, which is proved seriously limiting the expressivity and generalization. One potential remedy is incorporating the pre-trained knowledge within the "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2411.15633","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2024-11-23T19:10:32Z","cross_cats_sorted":[],"title_canon_sha256":"8562b9d00d4e88e6d8d323576b11263ac5229bdb152520a874df7728dfa341ae","abstract_canon_sha256":"439fd46a386a914a6a76f029ae6262bf358b60d175f9cd814cb6f494a0d828ed"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:37:42.503706Z","signature_b64":"sx7KwQ2rFUErVTBHUcN+0rQDeODr0/Y4n2PVRs7V5xjdovLFuPPUYjbwisk5W/yESWb+6CoBhYqyRRCDfOeQAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6ceb7d40a635cbbc046d40f9dbf791359730559d773de73f6faa62963527dde1","last_reissued_at":"2026-05-17T23:37:42.503015Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:37:42.503015Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Orthogonal Subspace Decomposition for Generalizable AI-Generated Image Detection","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Decomposing features via SVD into orthogonal parts lets detectors freeze general pre-trained knowledge and adapt only the rest to spot AI fakes without overfitting.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Baoyuan Wu, Chengchun Liu, Jiangming Wang, Ke-Yue Zhang, Li Yuan, Peng Jin, Shen Chen, Shouhong Ding, Taiping Yao, Zhiyuan Yan","submitted_at":"2024-11-23T19:10:32Z","abstract_excerpt":"AI-generated images (AIGIs), such as natural or face images, have become increasingly important yet challenging. In this paper, we start from a new perspective to excavate the reason behind the failure generalization in AIGI detection, named the \\textit{asymmetry phenomenon}, where a naively trained detector tends to favor overfitting to the limited and monotonous fake patterns, causing the feature space to become highly constrained and low-ranked, which is proved seriously limiting the expressivity and generalization. One potential remedy is incorporating the pre-trained knowledge within the "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By employing Singular Value Decomposition (SVD) to decompose the original feature space into two orthogonal subspaces and freezing the principal components while adapting only the remained components, we preserve the pre-trained knowledge while learning fake patterns, effectively minimizing overfitting and enhancing generalization.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The principal components obtained from SVD on features of pre-trained vision foundation models capture general knowledge that remains useful and orthogonal to the specific patterns needed for detecting fakes, such that freezing them does not remove information critical for the detection task.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Orthogonal subspace decomposition via SVD on vision foundation model features preserves high-rank pre-trained knowledge by freezing principal components and adapting residuals, reducing overfitting for better generalization in AI-generated image detection.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Decomposing features via SVD into orthogonal parts lets detectors freeze general pre-trained knowledge and adapt only the rest to spot AI fakes without overfitting.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"fb9fd55d714c7c73ea3b2f6e2ea9755c579554edd1109711046b82587e5bd71f"},"source":{"id":"2411.15633","kind":"arxiv","version":4},"verdict":{"id":"1f5ff1e7-c1cd-4cd0-9a72-573247869d30","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T23:20:52.690182Z","strongest_claim":"By employing Singular Value Decomposition (SVD) to decompose the original feature space into two orthogonal subspaces and freezing the principal components while adapting only the remained components, we preserve the pre-trained knowledge while learning fake patterns, effectively minimizing overfitting and enhancing generalization.","one_line_summary":"Orthogonal subspace decomposition via SVD on vision foundation model features preserves high-rank pre-trained knowledge by freezing principal components and adapting residuals, reducing overfitting for better generalization in AI-generated image detection.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The principal components obtained from SVD on features of pre-trained vision foundation models capture general knowledge that remains useful and orthogonal to the specific patterns needed for detecting fakes, such that freezing them does not remove information critical for the detection task.","pith_extraction_headline":"Decomposing features via SVD into orthogonal parts lets detectors freeze general pre-trained knowledge and adapt only the rest to spot AI fakes without overfitting."},"references":{"count":290,"sample":[{"doi":"","year":2022,"title":"Wukong, 2022. 5. In https://xihe.mindspore.cn/modelzoo/wukong, 2022. 5","work_id":"","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Brock, A. et al. Large scale gan training for high fidelity natural image synthesis. In ICLR, 2018 b","work_id":"","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"End-to-end reconstruction-classification learning for face forgery detection","work_id":"","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"What makes fake images detectable? understanding properties that generalize","work_id":"","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Drct: Diffusion reconstruction contrastive training towards universal detection of diffusion generated images","work_id":"","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":207,"snapshot_sha256":"d26ace6ba3622d8dd964c0505daa69045a38903242cd6d6af2f0a6bd2bea4755","internal_anchors":16},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2a965cf11ff36ab0957842805e72fe9e20f754411a4144cb0213f4147c92a42e"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2411.15633","created_at":"2026-05-17T23:37:42.503111+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.15633v4","created_at":"2026-05-17T23:37:42.503111+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.15633","created_at":"2026-05-17T23:37:42.503111+00:00"},{"alias_kind":"pith_short_12","alias_value":"NTVX2QFGGXF3","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"NTVX2QFGGXF3YBDN","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"NTVX2QFG","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":22,"internal_anchor_count":22,"sample":[{"citing_arxiv_id":"2503.21210","citing_title":"Toward Generalizable Forgery Detection and Reasoning","ref_index":85,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07074","citing_title":"Decoupling Semantics and Fingerprints: A Universal Representation for AI-Generated Image Detection","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2508.01994","citing_title":"Deeply Dual Supervised learning for melanoma recognition","ref_index":143,"is_internal_anchor":true},{"citing_arxiv_id":"2508.06805","citing_title":"Edge Detection for Organ Boundaries via Top Down Refinement and SubPixel Upsampling","ref_index":136,"is_internal_anchor":true},{"citing_arxiv_id":"2508.06816","citing_title":"DualResolution Residual Architecture with Artifact Suppression for Melanocytic Lesion Segmentation","ref_index":138,"is_internal_anchor":true},{"citing_arxiv_id":"2508.06819","citing_title":"VesselRW: Weakly Supervised Subcutaneous Vessel Segmentation via Learned Random Walk Propagation","ref_index":134,"is_internal_anchor":true},{"citing_arxiv_id":"2511.16136","citing_title":"How Noise Benefits AI-generated Image Detection","ref_index":74,"is_internal_anchor":true},{"citing_arxiv_id":"2512.12982","citing_title":"Scaling Up AI-Generated Image Detection with Generator-Aware Prototypes","ref_index":71,"is_internal_anchor":true},{"citing_arxiv_id":"2602.01738","citing_title":"Simplicity Prevails: The Emergence of Generalizable AIGI Detection in Visual Foundation Models","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2605.14486","citing_title":"Reduce the Artifacts Bias for More Generalizable AI-Generated Image Detection","ref_index":62,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03555","citing_title":"HEDGE: Heterogeneous Ensemble for Detection of AI-GEnerated Images in the Wild","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03558","citing_title":"LOGER: Local--Global Ensemble for Robust Deepfake Detection in the Wild","ref_index":58,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09296","citing_title":"Micro-Defects Expose Macro-Fakes: Detecting AI-Generated Images via Local Distributional Shifts","ref_index":50,"is_internal_anchor":true},{"citing_arxiv_id":"2604.25889","citing_title":"Robust Deepfake Detection: Mitigating Spatial Attention Drift via Calibrated Complementary Ensembles","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21478","citing_title":"Rethinking Cross-Domain Evaluation for Face Forgery Detection with Semantic Fine-grained Alignment and Mixture-of-Experts","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2604.12307","citing_title":"Boosting Robust AIGI Detection with LoRA-based Pairwise Training","ref_index":39,"is_internal_anchor":true},{"citing_arxiv_id":"2604.12353","citing_title":"Combating Pattern and Content Bias: Adversarial Feature Learning for Generalized AI-Generated Image Detection","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07074","citing_title":"Decoupling Semantics and Fingerprints: A Universal Representation for AI-Generated Image Detection","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2604.13660","citing_title":"VRAG-DFD: Verifiable Retrieval-Augmentation for MLLM-based Deepfake Detection","ref_index":50,"is_internal_anchor":true},{"citing_arxiv_id":"2604.16879","citing_title":"Adaptive Forensic Feature Refinement via Intrinsic Importance Perception","ref_index":56,"is_internal_anchor":true},{"citing_arxiv_id":"2604.17376","citing_title":"Towards Generalizable Deepfake Image Detection with Vision Transformers","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2605.04445","citing_title":"LEGO: LoRA-Enabled Generator-Oriented Framework for Synthetic Image Detection","ref_index":51,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW","json":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW.json","graph_json":"https://pith.science/api/pith-number/NTVX2QFGGXF3YBDNID45X54RGW/graph.json","events_json":"https://pith.science/api/pith-number/NTVX2QFGGXF3YBDNID45X54RGW/events.json","paper":"https://pith.science/paper/NTVX2QFG"},"agent_actions":{"view_html":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW","download_json":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW.json","view_paper":"https://pith.science/paper/NTVX2QFG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.15633&json=true","fetch_graph":"https://pith.science/api/pith-number/NTVX2QFGGXF3YBDNID45X54RGW/graph.json","fetch_events":"https://pith.science/api/pith-number/NTVX2QFGGXF3YBDNID45X54RGW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW/action/storage_attestation","attest_author":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW/action/author_attestation","sign_citation":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW/action/citation_signature","submit_replication":"https://pith.science/pith/NTVX2QFGGXF3YBDNID45X54RGW/action/replication_record"}},"created_at":"2026-05-17T23:37:42.503111+00:00","updated_at":"2026-05-17T23:37:42.503111+00:00"}