Research map / Vision-language and multimodal / Multimodal LLMs
MUSE-VL: Modeling Unified VLM through Semantic Discrete Encoding
Rongchang Xie, Chen Du, Ping Song, Chang Liu
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 |
|---|---|---|---|
| replace VQGAN tokenizer with semantic SDE tokenizer | Yi-1.5-9B VLM · MMBench/SEED/MMStar average | average accuracy: 34.6 -> 60.8 | Improved |
| use SDE tokenizer instead of SEED tokenizer | Yi-1.5-9B VLM · MMBench/SEED/MMStar average | average accuracy: 53.3 -> 60.8 | Improved |
| use SDE tokenizer instead of LaVIT tokenizer | Yi-1.5-9B VLM · MMBench/SEED/MMStar average | average accuracy: 54.4 -> 60.8 | Improved |
| add semantic branch (SigLIP feature reconstruction) to image-only tokenizer (ablation) | SDE tokenizer · MMB/SEED/MMStar average | average accuracy: 43.1 -> 63.6 | Improved |
| add semantic branch to image-only tokenizer (ablation) | SDE tokenizer · ImageNet val reconstruction | rFID: 2.63 -> 2.26 | Improved |
| add image reconstruction branch to semantic-only tokenizer (ablation) | SDE tokenizer · MMB/SEED/MMStar average | average accuracy: 62.7 -> 63.6 | Improved |
| increase input resolution from 256 to 384 (ablation) | MUSE-VL Yi-1.5-9B · MMB/SEED/MMStar average | average accuracy: 60.2 -> 63.3 | Improved |
| scale LLM from Yi-1.5-9B to Yi-1.5-34B (ablation) | MUSE-VL at 256 res · MMB/SEED/MMStar average | average accuracy: 60.2 -> 63.2 | Improved |
| scale LLM from Qwen-2.5-7B to Qwen-2.5-32B (ablation) | MUSE-VL at 256 res · MMB/SEED/MMStar average | average accuracy: 60.3 -> 63.7 | Improved |
| scale LLM from Qwen-2.5-7B to Qwen-2.5-32B (ablation) | MUSE-VL · SEED-Bench-img (ablation subset) | accuracy: 65.8 -> 65.7 | No clear effect |
| prompt rewriting for text-to-image generation | MUSE-VL 7B · GenEval | overall score: 0.53 -> 0.57 | Improved |
| scale MUSE-VL from 7B to 32B | MUSE-VL · Understanding average (7 benchmarks) | average accuracy: 63.6 -> 70.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- Janus — Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation (CVPR 2025) · better on 1 of 2 shared results
- Show-o — Show-o: One Single Transformer to Unify Multimodal Understanding and Generation (ICLR 2025) · 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
- SynerGen-VL: Towards Synergistic Image Understanding and Generation with Vision Experts and Token Folding (CVPR 2025 · similarity 0.91)
- Vision as a Dialect: Unifying Visual Understanding and Generation via Text-Aligned Representations (NeurIPS 2025 · similarity 0.91)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.90)
- Unified Language-Vision Pretraining in LLM with Dynamic Discrete Visual Tokenization (ICLR 2024 · similarity 0.90)
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models (NeurIPS 2025 · similarity 0.90)
- UniTok: a Unified Tokenizer for Visual Generation and Understanding (NeurIPS 2025 · similarity 0.90)
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.
