{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VKMDCI5UZDSZWFNCV7XDXBOITO","short_pith_number":"pith:VKMDCI5U","schema_version":"1.0","canonical_sha256":"aa983123b4c8e59b15a2afee3b85c89bb8d2eb712c9fc07abeba187b484624ad","source":{"kind":"arxiv","id":"2503.11905","version":1},"attestation_state":"computed","paper":{"title":"Upcycling Text-to-Image Diffusion Models for Multi-Task Capabilities","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Abhinav Mehrotra, Alberto Gil Ramos, Luca Morreale, Malcolm Chadwick, Mehdi Noroozi, Ruchika Chavhan, Sourav Bhattacharya","submitted_at":"2025-03-14T22:19:20Z","abstract_excerpt":"Text-to-image synthesis has witnessed remarkable advancements in recent years. Many attempts have been made to adopt text-to-image models to support multiple tasks. However, existing approaches typically require resource-intensive re-training or additional parameters to accommodate for the new tasks, which makes the model inefficient for on-device deployment. We propose Multi-Task Upcycling (MTU), a simple yet effective recipe that extends the capabilities of a pre-trained text-to-image diffusion model to support a variety of image-to-image generation tasks. MTU replaces Feed-Forward Network ("},"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":"2503.11905","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-03-14T22:19:20Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"59206324b3f3bbbd394c1cee795a10eb453f8d7c4f16a77801ca3c7153dc934b","abstract_canon_sha256":"b4a1b5df6e09f817dfb9206f79ff5a2e9e714954de97b5ff6fec53d60ed9ac4a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:32:01.747734Z","signature_b64":"HGC/+TxUD5ygRUor5lg3hVAJNM+DUdO2djdKmmitBPrTgUkLz1DcZaRqddVvnFjnYWbfZ40zT6+CCYSGlA4fCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aa983123b4c8e59b15a2afee3b85c89bb8d2eb712c9fc07abeba187b484624ad","last_reissued_at":"2026-07-05T10:32:01.746941Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:32:01.746941Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Upcycling Text-to-Image Diffusion Models for Multi-Task Capabilities","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Abhinav Mehrotra, Alberto Gil Ramos, Luca Morreale, Malcolm Chadwick, Mehdi Noroozi, Ruchika Chavhan, Sourav Bhattacharya","submitted_at":"2025-03-14T22:19:20Z","abstract_excerpt":"Text-to-image synthesis has witnessed remarkable advancements in recent years. Many attempts have been made to adopt text-to-image models to support multiple tasks. However, existing approaches typically require resource-intensive re-training or additional parameters to accommodate for the new tasks, which makes the model inefficient for on-device deployment. We propose Multi-Task Upcycling (MTU), a simple yet effective recipe that extends the capabilities of a pre-trained text-to-image diffusion model to support a variety of image-to-image generation tasks. MTU replaces Feed-Forward Network ("},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.11905","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2503.11905/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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":"2503.11905","created_at":"2026-07-05T10:32:01.747055+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.11905v1","created_at":"2026-07-05T10:32:01.747055+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.11905","created_at":"2026-07-05T10:32:01.747055+00:00"},{"alias_kind":"pith_short_12","alias_value":"VKMDCI5UZDSZ","created_at":"2026-07-05T10:32:01.747055+00:00"},{"alias_kind":"pith_short_16","alias_value":"VKMDCI5UZDSZWFNC","created_at":"2026-07-05T10:32:01.747055+00:00"},{"alias_kind":"pith_short_8","alias_value":"VKMDCI5U","created_at":"2026-07-05T10:32:01.747055+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2411.18823","citing_title":"Multi-Task Label Discovery via Hierarchical Task Tokens for Partially Annotated Dense Predictions","ref_index":7,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO","json":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO.json","graph_json":"https://pith.science/api/pith-number/VKMDCI5UZDSZWFNCV7XDXBOITO/graph.json","events_json":"https://pith.science/api/pith-number/VKMDCI5UZDSZWFNCV7XDXBOITO/events.json","paper":"https://pith.science/paper/VKMDCI5U"},"agent_actions":{"view_html":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO","download_json":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO.json","view_paper":"https://pith.science/paper/VKMDCI5U","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.11905&json=true","fetch_graph":"https://pith.science/api/pith-number/VKMDCI5UZDSZWFNCV7XDXBOITO/graph.json","fetch_events":"https://pith.science/api/pith-number/VKMDCI5UZDSZWFNCV7XDXBOITO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO/action/storage_attestation","attest_author":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO/action/author_attestation","sign_citation":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO/action/citation_signature","submit_replication":"https://pith.science/pith/VKMDCI5UZDSZWFNCV7XDXBOITO/action/replication_record"}},"created_at":"2026-07-05T10:32:01.747055+00:00","updated_at":"2026-07-05T10:32:01.747055+00:00"}