Research map / Image generation / Text-to-image diffusion
Explain Me the Painting: Multi-Topic Knowledgeable Art Description Generation
Zechen Bai, Yuta Nakashima, Noa Garcia
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add topic-parallel decoder instead of topic-conditional decoder (ablation) | Ours · SemArt test | BLEU-4: 0.9 -> 8.8 | Improved |
| add masked-sentence generation and knowledge filling to the topic decoder (vs MScap stylized captioning) | Ours (Conditional decoder) · SemArt test | BLEU-4: 0.4 -> 8.8 | Improved |
| add retrieved Wikipedia knowledge and topic decoder over SAT baseline | SAT · SemArt human evaluation | Relevance: 1.94 -> 2.29 | Improved |
| add retrieved Wikipedia knowledge and topic decoder over SAT baseline | SAT · SemArt human evaluation | Veracity: 1.30 -> 1.56 | Improved |
| add retrieved Wikipedia knowledge and topic decoder over SAT baseline | SAT · SemArt human evaluation | Understandability: 3.62 -> 2.71 | Hurt |
| add topic decoder to SAT baseline | SAT · SemArt human evaluation | Form existence: 0.05 -> 0.33 | Improved |
| use original SemArt comment (oracle knowledge) instead of retrieved Wikipedia (ablation) | Ours · SemArt human evaluation | Veracity: 1.56 -> 1.72 | Improved |
| use retrieved SemArt comments instead of retrieved Wikipedia as knowledge source (ablation) | Ours · SemArt human evaluation | Relevance: 2.29 -> 2.02 | Hurt |
| use semi-supervised Guided-LDA topic labels instead of unsupervised LDA (ablation) | topic decoder · SemArt topic labels | topic label accuracy: 43.3 -> 51.6 | Improved |
| add generated masked sentences to the retrieval query (ablation) | knowledge retriever · SemArt knowledge retrieval (150 paintings) | Recall@10 (all articles): 45.5 -> 17.5 | Hurt |
| add generated masked sentences to the retrieval query (ablation) | knowledge retriever · SemArt knowledge retrieval (150 paintings) | Recall@1 (all articles): 13.8 -> 5.0 | Hurt |
| add parallel-decoder multi-topic method over classic captioning | NIC/SAT/Att2in · SemArt test | CIDEr: 38.6 -> 9.1 | Hurt |
Closest papers by meaning
- Bridging Cognitive Gap: Hierarchical Description Learning for Artistic Image Aesthetics Assessment (AAAI 2026 · similarity 0.82)
- PoSh: Using Scene Graphs to Guide LLMs-as-a-Judge for Detailed Image Descriptions (ICLR 2026 · similarity 0.82)
- Text-to-Image Generation for Abstract Concepts (AAAI 2024 · similarity 0.82)
- Stylebreeder: Exploring and Democratizing Artistic Styles through Text-to-Image Models (NeurIPS 2024 · similarity 0.81)
- APDDv2: Aesthetics of Paintings and Drawings Dataset with Artist Labeled Scores and Comments (NeurIPS 2024 · similarity 0.81)
- The Photographer's Eye: Teaching Multimodal Large Language Models to See, and Critique Like Photographers (CVPR 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
