AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

LMM4LMM: Benchmarking and Evaluating Large-multimodal Image Generation with LMMs

Jiarui Wang, Huiyu Duan, Yu Zhao, Juntong Wang, Guangtao Zhai, Xiongkuo Min

ICCV 2025 · Image generation › Text-to-image diffusion · approach: T2I evaluation and controllable adapters · 42 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 LoRA fine-tuning to both vision encoder and LLM (ablation)LMM4LMM (InternVL2.5-8B) · EvalMi-50K perceptionSRCC: 0.828 -> 0.886Improved
add text-defined quality-level initialization (ablation)LMM4LMM · EvalMi-50K perceptionSRCC: 0.865 -> 0.886Improved
add LoRA to LLM only (vs full LoRA on both) (ablation)LMM4LMM · EvalMi-50K perceptionSRCC: 0.872 -> 0.886 when also adding vision LoRAImproved
scale backbone from InternVL2.5 8B to 26B (ablation)LMM4LMM · EvalMi-50K perceptionSRCC: 0.886 -> 0.882No clear effect
swap InternVL2.5-8B backbone for Qwen2.5-VL 8B (ablation)LMM4LMM · EvalMi-50K perceptionSRCC: 0.886 -> 0.699Hurt
swap backbone for Llama3.2-Vision 11B (ablation)LMM4LMM · EvalMi-50K QA accuracyQA accuracy: 87.9% -> 78.1%Hurt
fine-tune off-the-shelf InternVL2.5-style LMMs with the paper's recipe vs zero-shot LMMsLMMs for scoring · EvalMi-50K perceptionSRCC: 0.2597 -> 0.8863Improved
train on EvalMuse-40K (coarse MOS) instead of EvalMi-50K (ablation)LMM4LMM · GenAI-Bench zero-shotSRCC: 0.7082 -> 0.6523Hurt
use LMM4LMM instead of Q-Align for T2I model rankingperception score ranking · EvalMi-50K T2I modelsSRCC to human: 0.940 -> 0.979Improved
LMM4LMM vs FGA-BLIP2 on EvalMuse-40K in-domainLMM4LMM (trained on EvalMi) · EvalMuse-40KSRCC: 0.6560 vs 0.7723*Mixed

Later papers that compared with this method

1 later accepted paper reports LMM4LMM as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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