{"paper":{"title":"Skill-Aligned Annotation for Reliable Evaluation in Text-to-Image Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Annotation strategies tailored to each evaluation skill produce more consistent signals and higher agreement than uniform scales across all skills in text-to-image generation.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Abdelrahman Eldesokey, Ahmad Sait, Ansar Khangeldin, Bernard Ghanem, Karen Sanchez, Merey Ramazanova, Tong Zhang","submitted_at":"2026-05-13T09:14:31Z","abstract_excerpt":"Text-to-image (T2I) generation has advanced rapidly, making reliable evaluation critical as performance differences between models narrow. Existing evaluation practices typically apply uniform annotation mechanisms, such as Likert-scale or binary question answering (BQA), across heterogeneous evaluation skills, despite fundamental differences in their nature. In this work, we revisit T2I evaluation through the lens of skill-aligned annotation, where annotation strategies reflect the underlying characteristics of each evaluation skill. We systematically compare skill-aligned annotation against "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"skill-aligned annotation produces more consistent evaluation signals, with higher inter-annotator agreement and improved stability across models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the chosen skill-aligned annotation strategies are fundamentally better suited to each skill's nature and that the uniform baselines provide a fair comparison without confounding factors in skill selection or annotation design.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Skill-aligned annotation improves inter-annotator agreement and evaluation stability in text-to-image generation compared to uniform annotation baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Annotation strategies tailored to each evaluation skill produce more consistent signals and higher agreement than uniform scales across all skills in text-to-image generation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"dda269bc01196853103c04ecda507bc591c2b09b6bbbf3a0e0a95ebf9b138334"},"source":{"id":"2605.13223","kind":"arxiv","version":1},"verdict":{"id":"dfe24b60-f4ea-4214-9310-a8a75ad93df6","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:09:22.571008Z","strongest_claim":"skill-aligned annotation produces more consistent evaluation signals, with higher inter-annotator agreement and improved stability across models.","one_line_summary":"Skill-aligned annotation improves inter-annotator agreement and evaluation stability in text-to-image generation compared to uniform annotation baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the chosen skill-aligned annotation strategies are fundamentally better suited to each skill's nature and that the uniform baselines provide a fair comparison without confounding factors in skill selection or annotation design.","pith_extraction_headline":"Annotation strategies tailored to each evaluation skill produce more consistent signals and higher agreement than uniform scales across all skills in text-to-image generation."},"references":{"count":42,"sample":[{"doi":"","year":2025,"title":"Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer","work_id":"f1080a62-48e1-4255-b023-7556be57370d","ref_index":1,"cited_arxiv_id":"2511.22699","is_internal_anchor":true},{"doi":"","year":2024,"title":"J. Chen, J. YU, C. GE, L. Y ao, E. Xie, Z. Wang, J. Kwok, P . Luo, H. Lu, and Z. Li. Pixart- $\\alpha$: Fast training of diffusion transformer for photorealistic text-to-image synthesis. In The Twelfth","work_id":"f7a143e0-8789-4955-915c-77dd726ffe92","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"J. Cho, Y . Hu, J. M. Baldridge, R. Garg, P . Anderson, R. Krishna, M. Bansal, J. Pont-Tuset, and S. Wang. Davidsonian scene graph: Improving reliability in fine-grained evaluation for text-to-image ge","work_id":"84a58b20-1f7c-4871-971b-da2c2932397d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding","work_id":"f0dc2969-bd90-4a5d-81df-d557352690b6","ref_index":4,"cited_arxiv_id":"2601.10611","is_internal_anchor":true},{"doi":"","year":2025,"title":"G. DeepMind. Nano-banana. https://deepmind.google/models/gemini-image/ flash/, 2025","work_id":"0a43bc6f-7c1c-4a5d-93d8-001740cc1b6c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":42,"snapshot_sha256":"a4608d45b293e31e6ebd7b7a1ff4fe88c266a82c4f899fbdd4ba99569b366297","internal_anchors":5},"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"}