{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2017:BWYZ5PAJSARALRMON6YFVTQVFD","short_pith_number":"pith:BWYZ5PAJ","schema_version":"1.0","canonical_sha256":"0db19ebc09902205c58e6fb05ace1528e458bac2c4ee722a55e0eadb6e4143e3","source":{"kind":"arxiv","id":"1707.05612","version":4},"attestation_state":"computed","paper":{"title":"VSE++: Improving Visual-Semantic Embeddings with Hard Negatives","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"David J. Fleet, Fartash Faghri, Jamie Ryan Kiros, Sanja Fidler","submitted_at":"2017-07-18T13:51:32Z","abstract_excerpt":"We present a new technique for learning visual-semantic embeddings for cross-modal retrieval. Inspired by hard negative mining, the use of hard negatives in structured prediction, and ranking loss functions, we introduce a simple change to common loss functions used for multi-modal embeddings. That, combined with fine-tuning and use of augmented data, yields significant gains in retrieval performance. We showcase our approach, VSE++, on MS-COCO and Flickr30K datasets, using ablation studies and comparisons with existing methods. On MS-COCO our approach outperforms state-of-the-art methods by 8"},"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":false,"formal_links_present":false},"canonical_record":{"source":{"id":"1707.05612","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2017-07-18T13:51:32Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"f6aafd43e8d5afb01ef53afc75c809df67dd98c8f119ec9bb3133aea7fa087da","abstract_canon_sha256":"9c5c7c56ce1b9b3416611767ded9ac8b8bcea192d6b3f87fc6f3af7e3b75d59d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T00:09:36.556546Z","signature_b64":"sGbUbpplgZnkwHisfhB8r06h5thRZxl2WXZhvqRtE8AQOgtmjnJnFVUNQNrf4ROflMDrqcfY2wnGYbvGS/5VAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0db19ebc09902205c58e6fb05ace1528e458bac2c4ee722a55e0eadb6e4143e3","last_reissued_at":"2026-05-18T00:09:36.556030Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T00:09:36.556030Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"VSE++: Improving Visual-Semantic Embeddings with Hard Negatives","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"David J. Fleet, Fartash Faghri, Jamie Ryan Kiros, Sanja Fidler","submitted_at":"2017-07-18T13:51:32Z","abstract_excerpt":"We present a new technique for learning visual-semantic embeddings for cross-modal retrieval. Inspired by hard negative mining, the use of hard negatives in structured prediction, and ranking loss functions, we introduce a simple change to common loss functions used for multi-modal embeddings. That, combined with fine-tuning and use of augmented data, yields significant gains in retrieval performance. We showcase our approach, VSE++, on MS-COCO and Flickr30K datasets, using ablation studies and comparisons with existing methods. On MS-COCO our approach outperforms state-of-the-art methods by 8"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1707.05612","kind":"arxiv","version":4},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"1707.05612","created_at":"2026-05-18T00:09:36.556122+00:00"},{"alias_kind":"arxiv_version","alias_value":"1707.05612v4","created_at":"2026-05-18T00:09:36.556122+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1707.05612","created_at":"2026-05-18T00:09:36.556122+00:00"},{"alias_kind":"pith_short_12","alias_value":"BWYZ5PAJSARA","created_at":"2026-05-18T12:31:08.081275+00:00"},{"alias_kind":"pith_short_16","alias_value":"BWYZ5PAJSARALRMO","created_at":"2026-05-18T12:31:08.081275+00:00"},{"alias_kind":"pith_short_8","alias_value":"BWYZ5PAJ","created_at":"2026-05-18T12:31:08.081275+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.15723","citing_title":"GOMA: Toward Structure-Driven Multimodal Alignment from a Graph Signal Smoothing Perspective","ref_index":59,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17766","citing_title":"LatentUMM: Dual Latent Alignment for Unified Multimodal Models","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2305.16355","citing_title":"PandaGPT: One Model To Instruction-Follow Them All","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2604.22885","citing_title":"Federated Cross-Modal Retrieval with Missing Modalities via Semantic Routing and Adapter Personalization","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15735","citing_title":"Sketch and Text Synergy: Fusing Structural Contours and Descriptive Attributes for Fine-Grained Image Retrieval","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD","json":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD.json","graph_json":"https://pith.science/api/pith-number/BWYZ5PAJSARALRMON6YFVTQVFD/graph.json","events_json":"https://pith.science/api/pith-number/BWYZ5PAJSARALRMON6YFVTQVFD/events.json","paper":"https://pith.science/paper/BWYZ5PAJ"},"agent_actions":{"view_html":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD","download_json":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD.json","view_paper":"https://pith.science/paper/BWYZ5PAJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1707.05612&json=true","fetch_graph":"https://pith.science/api/pith-number/BWYZ5PAJSARALRMON6YFVTQVFD/graph.json","fetch_events":"https://pith.science/api/pith-number/BWYZ5PAJSARALRMON6YFVTQVFD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD/action/storage_attestation","attest_author":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD/action/author_attestation","sign_citation":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD/action/citation_signature","submit_replication":"https://pith.science/pith/BWYZ5PAJSARALRMON6YFVTQVFD/action/replication_record"}},"created_at":"2026-05-18T00:09:36.556122+00:00","updated_at":"2026-05-18T00:09:36.556122+00:00"}