AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

PUMA: Empowering Unified MLLM with Multi-granular Visual Generation

Rongyao Fang, Chengqi Duan, Kun Wang, Hao Li, Linjiang Huang, Hao Tian, Xingyu Zeng, Rui Zhao, Jifeng Dai, Hongsheng Li, Xihui Liu

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Instruction-tuned multimodal assistants · 24 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
use f0 (256 tokens) instead of f4 (1 token) features for decoding (ablation)PUMA · ImageNet val reconstructionPSNR_r: 10.76 -> 18.16Improved
use f4 (1 token) instead of f0 features for decoding (ablation)PUMA · ImageNet val decoding diversityLPIPS_d: 0.1559 -> 0.5751Improved
use f3 scale instead of f4 scale (ablation)PUMA · MSCOCO 30K text-to-imageCLIP-I: 0.699 -> 0.703Improved
use f3 scale instead of f4 scale (ablation)PUMA · MSCOCO 30K text-to-imageLPIPS_d: 0.613 -> 0.558Hurt
use f0 scale instead of f2 for text-to-image (ablation)PUMA · MSCOCO 30K text-to-imageCLIP-I: 0.703 -> 0.621Hurt
select best of 5 scale outputs (5-scale Max) (ablation)PUMA · MSCOCO 30K text-to-imageCLIP-I: 0.703 -> 0.736Improved
use f0 instead of f1 scale features (ablation)PUMA · Emu-Edit testDINO: 0.679 -> 0.784Improved
use f0 instead of f1 scale features (ablation)PUMA · Emu-Edit testCLIP-I: 0.802 -> 0.840Improved
use f0 instead of f1 scale features (ablation)PUMA · Emu-Edit testCLIP-T: 0.258 -> 0.264No clear effect
use all 5 scales f4-f0 as understanding input instead of f4 only (ablation)PUMA encoder · MMBenchscore: 56.8 -> 65.1Improved
use f0 (256 tokens) instead of f2 (16 tokens) as understanding input (ablation)PUMA encoder · GQAscore: 46.6 -> 58.8Improved
use f4-f0 (341 tokens) instead of f0 (256 tokens) as understanding input (ablation)PUMA encoder · GQAscore: 58.8 -> 61.0Improved

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