Research map / Vision-language and multimodal / Human and embodied modeling
MetaCLUE: Towards Comprehensive Visual Metaphors Research
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use ViT-B/16 rather than EfficientNet-B0 for metaphor classification | metaphor classification · MetaCLUE metaphor vs. symbolic negatives test | accuracy: 49.67 -> 66.98 | Helped |
| Use ViT-B/16 rather than EfficientNet-B0 for metaphor classification | metaphor classification · MetaCLUE metaphor vs. literal negatives test | accuracy: 50.30 -> 81.24 | Helped |
| Fine-tune CLIP on metaphor annotations | CLIP (ViT-L/14) retrieval · MetaCLUE retrieval, random negatives | p@1: 76.66 -> 81.75 | Helped |
| Use hard negatives instead of random negatives | CLIP retrieval models · MetaCLUE retrieval | p@1: 76.66 -> 39.86 | Hurt |
| Replace the primary concept rather than the secondary concept in hard negatives | CLIP (ViT-L/14) retrieval · MetaCLUE retrieval, hard negatives | p@1: 39.86 -> 51.75 | Helped |
| Fine-tune literal image-captioning PaLI for metaphor captioning | PaLI-17B · MetaCLUE metaphor captioning, whole caption | exact match accuracy | Hurt |
| Fine-tune literal VQA PaLI for metaphor VQA | PaLI-17B · MetaCLUE metaphor VQA, all questions | accuracy | Hurt |
| Ask about the secondary concept rather than the primary concept | PaLI-17B metaphor VQA · MetaCLUE metaphor VQA | accuracy: 21.5% -> 12.8% | Hurt |
| Localize secondary concepts rather than primary concepts | CLIP-based phrase localization · MetaCLUE test split | mAP50: 33.22 -> 43.54 | Helped |
| Fine-tune Stable Diffusion on the MetaCLUE train split (ablation) | Stable Diffusion · MetaCLUE metaphor generation | FID: 161.6 -> 154.3 | Helped |
| Fine-tune Stable Diffusion on the MetaCLUE train split (ablation) | Stable Diffusion · MetaCLUE metaphor generation | CLIP Similarity: 30.8 -> 32.0 | Helped |
| Fine-tune Stable Diffusion on the MetaCLUE train split (ablation) | Stable Diffusion · human preference for metaphor generation | user preference | Hurt |
Closest papers by meaning
- When Visualizing is the First Step to Reasoning: MIRA, a Benchmark for Visual Chain-of-Thought (CVPR 2026 · similarity 0.87)
- ARC Is a Vision Problem! (CVPR 2026 · similarity 0.86)
- Visual Thoughts: A Unified Perspective of Understanding Multimodal Chain-of-Thought (NeurIPS 2025 · similarity 0.86)
- Retrieving Counterfactuals Improves Visual In-Context Learning (CVPR 2026 · similarity 0.86)
- Text-to-Image Generation for Abstract Concepts (AAAI 2024 · similarity 0.86)
- SANEval: Open-Vocabulary Compositional Benchmarks with Failure-mode Diagnosis (NeurIPS 2026 · similarity 0.85)
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.