Research map / Vision-language and multimodal / Multimodal LLMs
HumorDB: Can AI understand graphical humor?
Vedaant V Jain, Gabriel Kreiman, Felipe dos Santos Alves Feitosa
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 |
|---|---|---|---|
| fine-tune LLaVA instead of zero-shot | LLaVA · HumorDB Range task (testAllSet) | RMSE: 2.95 -> 1.70 | Improved |
| add human keywords prefix when fine-tuning (ablation) | LLaVA · HumorDB Range task (testAllSet) | RMSE: 1.70 -> 1.68 | Improved |
| add human keywords prefix when fine-tuning (ablation) | BLIP · HumorDB Range task (testAllSet) | RMSE: 1.94 -> 1.92 | No clear effect |
| add human keywords prefix when fine-tuning (ablation) | LLaVA · HumorDB Binary task (Photoshopped images) | accuracy (%): 76 -> 83 | Improved |
| add human keywords prefix when fine-tuning (ablation) | LLaVA · HumorDB Binary task (Photos images) | accuracy (%): 72 -> 79 | Improved |
| fine-tune LLaVA instead of zero-shot | LLaVA · HumorDB Binary task (Cartoons images) | accuracy (%): 52 -> 65 | Improved |
| add pretraining to vision models | vision-only models · HumorDB Binary task | accuracy (%) | Improved |
| restrict evaluation to paired images only | vision models · HumorDB Binary task | accuracy (%) | Hurt |
| evaluate only on original images, excluding modified images | GPT-4o · HumorDB Binary task | accuracy (%): 10% improvement | Improved |
| sketch images versus other image types | all models · HumorDB Binary task (Sketches images) | accuracy (%): 47 - 54 | Hurt |
| replace ViT-Huge by ViTG-14 (larger model) | vision-only fine-tuned · HumorDB Range task (testAllSet) | RMSE: 1.88 -> 1.76 | Improved |
| use large VLM Gemini-Flash zero-shot instead of GPT-4o zero-shot | zero-shot VLM · HumorDB Range task (testAllSet) | RMSE: 2.61 -> 2.06 | Improved |
Closest papers by meaning
- Cracking the Code of Juxtaposition: Can AI Models Understand the Humorous Contradictions (NeurIPS 2024 · similarity 0.89)
- HUMORCHAIN: Theory-Guided Multi-Stage Reasoning for Interpretable Multimodal Humor Generation (CVPR 2026 · similarity 0.89)
- From Perception to Punchline: Empowering VLM with the Art of In-the-wild Memes (NeurIPS 2026 · similarity 0.88)
- Humor in AI: Massive Scale Crowd-Sourced Preferences and Benchmarks for Cartoon Captioning (NeurIPS 2024 · similarity 0.88)
- SatireDecoder: Visual Cascaded Decoupling for Enhancing Satirical Image Comprehension (AAAI 2026 · similarity 0.86)
- ConTextual: Evaluating Context-Sensitive Text-Rich Visual Reasoning in Large Multimodal Models (ICML 2024 · 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.
