Research map / Vision-language and multimodal / Multimodal LLMs
Engage for All: Making Ordinary Image Descriptions Appealing Again!
Yuyan Chen, Yifan Jiang, Li Zhou, Jinghan Cao, Yu Guan, Ming Yang, Qingpei Guo
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 first-stage instruction tuning (QFormer alignment with frozen encoder) (ablation) | CharmNet · AppealImage | average score: 41.2 -> 45.0 | Improved |
| add heuristic active learning on hard samples (ablation) | CharmNet · AppealImage | average score: 37.6 -> 45.0 | Improved |
| add heuristic active learning on hard samples (ablation) | CharmNet · AppealImage generation | human appeal score: 3.5 -> 4.4 | Improved |
| replace heuristic active learning with DPO reinforcement learning (ablation) | CharmNet · AppealImage | average score: 45.0 -> 43.8 | No clear effect |
| add multi-task learning instead of generation-only training (ablation) | CharmNet · AppealImage generation | human appeal score: 4.2 -> 4.4 | Improved |
| increase hard-sample proportion D1 in active learning beyond 30% (ablation) | CharmNet · AppealImage | performance | Mixed |
| harder candidate descriptions (ablation) | CharmNet · AppealImage discrimination/selection/ranking | accuracy | Hurt |
| CharmNet versus GPT-4o | CharmNet · AppealImage discrimination | accuracy: 76.4 -> 77.5 | Improved |
| CharmNet versus best generation baseline LLaVA-1.6 on global perspective | CharmNet · AppealImage DSRG global | score: 49.5 -> 55.5 | Improved |
| CharmNet versus human-written descriptions | CharmNet · AppealImage global generation | average machine score: 48.6 -> 52.7 | Improved |
| CharmNet versus human-written description time | CharmNet · AppealImage global generation | seconds per description: 30.5 -> 3.2 | Improved |
| CharmNet versus CLoT tuned on AppealImage (ablation) | CharmNet · AppealImage generation | referee score: 0.34 -> 0.38 | Improved |
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.
- BLIP-2 — BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (ICML 2022) · better on 1 of 1 shared result
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
- From Perception to Punchline: Empowering VLM with the Art of In-the-wild Memes (NeurIPS 2026 · similarity 0.85)
- MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models (ICLR 2024 · similarity 0.85)
- HUMORCHAIN: Theory-Guided Multi-Stage Reasoning for Interpretable Multimodal Humor Generation (CVPR 2026 · similarity 0.84)
- Elucidating the design space of language models for image generation (ICML 2025 · similarity 0.84)
- Humor in AI: Massive Scale Crowd-Sourced Preferences and Benchmarks for Cartoon Captioning (NeurIPS 2024 · similarity 0.84)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · 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.
