AI Paper Atlas

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

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Interleaved image-text generation with MLLMs · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
add first-stage instruction tuning (QFormer alignment with frozen encoder) (ablation)CharmNet · AppealImageaverage score: 41.2 -> 45.0Improved
add heuristic active learning on hard samples (ablation)CharmNet · AppealImageaverage score: 37.6 -> 45.0Improved
add heuristic active learning on hard samples (ablation)CharmNet · AppealImage generationhuman appeal score: 3.5 -> 4.4Improved
replace heuristic active learning with DPO reinforcement learning (ablation)CharmNet · AppealImageaverage score: 45.0 -> 43.8No clear effect
add multi-task learning instead of generation-only training (ablation)CharmNet · AppealImage generationhuman appeal score: 4.2 -> 4.4Improved
increase hard-sample proportion D1 in active learning beyond 30% (ablation)CharmNet · AppealImageperformanceMixed
harder candidate descriptions (ablation)CharmNet · AppealImage discrimination/selection/rankingaccuracyHurt
CharmNet versus GPT-4oCharmNet · AppealImage discriminationaccuracy: 76.4 -> 77.5Improved
CharmNet versus best generation baseline LLaVA-1.6 on global perspectiveCharmNet · AppealImage DSRG globalscore: 49.5 -> 55.5Improved
CharmNet versus human-written descriptionsCharmNet · AppealImage global generationaverage machine score: 48.6 -> 52.7Improved
CharmNet versus human-written description timeCharmNet · AppealImage global generationseconds per description: 30.5 -> 3.2Improved
CharmNet versus CLoT tuned on AppealImage (ablation)CharmNet · AppealImage generationreferee score: 0.34 -> 0.38Improved

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy