Research map / Vision-language and multimodal / Multimodal LLMs
MetaGPT: A Large Vision-Language Model for Meme Metaphor Understanding
Bo Xu, Chenyuan Wang, Xinyu Chen, Hongfei Lin, Feng Xia
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces MetaGPT, described as the first vision-language model for meme metaphor understanding, aimed at recognizing and comprehending metaphors in memes without manual annotation.
Problem
Metaphor understanding in memes relies heavily on manual annotation, and mainstream vision-language models still struggle to recognize and comprehend meme metaphors.
Approach
A vision-language model dedicated to meme metaphor understanding, intended to reduce dependence on manual metaphor annotations. Details beyond the abstract are not available.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add OCR meme text as extra input (ablation) | MetaGPT · MEMECAP metaphor domain extraction (tau=0.5) | macro F1: 29.13 -> 26.44 | Hurt |
| replace inpainted patches with segmented original image regions (ablation) | MetaGPT · MEMECAP metaphor domain extraction (tau=0.5) | macro F1: 29.13 -> 27.46 | Hurt |
| add shallow and mid-level CLIP layer features (ablation) | MetaGPT · MEMECAP metaphor domain extraction (tau=0.5) | macro F1: 29.13 -> 28.33 / 28.63 / 28.03 | Hurt |
| add shallow and mid-level CLIP layer features (ablation) | MetaGPT · MET-Meme metaphor detection | weighted F1: 87.33 -> 86.54 / 86.83 / 86.96 | No clear effect |
| add OCR meme text as extra input (ablation) | MetaGPT · MEMECAP meme interpretation | average of 8 metrics: 55.89 -> 51.36 | Hurt |
| add MetaClean module (inpainting plus quadrant patches) (ablation) | MetaGPT · MEMECAP metaphor domain extraction (tau=0.5) | macro F1 | Improved |
| add MetaClean module (ablation) | MetaGPT · MET-Meme metaphor detection | F1 | No clear effect |
| increase MUnd training data ratio (ablation) | MetaGPT · all three tasks | task metrics | Improved |
| fine-tune LLaVA-v1.5-7B on MUnd with MetaClean vs larger baselines | MetaGPT · MET-Meme metaphor detection | weighted F1: 68.53 -> 87.33 | Improved |
| scale model parameters without metaphor training | LLaVA-v1.6 / CogVLM · MET-Meme metaphor detection | weighted F1: 19.19 and 12.18 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CogVLM — CogVLM: Visual Expert for Pretrained Language Models (NeurIPS 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Multi-Granular Multimodal Clue Fusion for Meme Understanding (AAAI 2025 · similarity 0.89)
- CAMEL: Capturing Metaphorical Alignment with Context Disentangling for Multimodal Emotion Recognition (AAAI 2024 · similarity 0.86)
- ViMU: Benchmarking Video Metaphorical Understanding (NeurIPS 2026 · similarity 0.85)
- From Perception to Punchline: Empowering VLM with the Art of In-the-wild Memes (NeurIPS 2026 · similarity 0.84)
- ConTextual: Evaluating Context-Sensitive Text-Rich Visual Reasoning in Large Multimodal Models (ICML 2024 · similarity 0.84)
- HUMORCHAIN: Theory-Guided Multi-Stage Reasoning for Interpretable Multimodal Humor Generation (CVPR 2026 · similarity 0.84)
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.
