Research map / Vision-language and multimodal / Human and embodied modeling
Unified Multimodal Models as Auto-Encoders
Zhiyuan Yan, Kaiqing Lin, Zongjian Li, Junyan Ye, Hui Han, Haochen Wang, Zhendong Wang, Bin Lin, Hao Li, Xinyan Xiao, Jingdong Wang, Haifeng Wang and 1 more
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 |
|---|---|---|---|
| apply Unified-GRPO reconstructive RL post-training (ablation) | UniWorld · GenEval | overall: 84.0 -> 89.0 | Improved |
| apply Unified-GRPO (ablation) | UniWorld · DPG-Bench | overall: 81.2 -> 86.4 | Improved |
| apply Unified-GRPO (ablation) | UniWorld · Unified-Score | unified score: 79.0 -> 86.1 | Improved |
| apply Unified-GRPO (ablation) | UniWorld · MMMU | accuracy: 58.6 -> 58.2 | No clear effect |
| apply Unified-GRPO (ablation) | UniWorld · MMBench | accuracy: 83.5 -> 84.8 | Improved |
| apply Unified-GRPO (ablation) | Janus-Pro · GenEval | overall: 80.0 -> 84.3 | Improved |
| apply Unified-GRPO (ablation) | Janus-Pro · Unified-Score | unified score: 82.8 -> 89.1 | Improved |
| apply Unified-GRPO to Qwen-2.5-VL-3B understanding model (ablation) | Qwen-2.5-VL-3B · MMT-Bench overall | accuracy: 56.3 -> 56.5 | No clear effect |
| apply Unified-GRPO to Qwen-2.5-VL-3B (ablation) | Qwen-2.5-VL-3B · MMT-Bench fine-grained perception | accuracy: 32.5 -> 56.9 | Improved |
| apply Unified-GRPO to Qwen-2.5-VL-3B (ablation) | Qwen-2.5-VL-3B · MMT-Bench Small Object Detection | accuracy: 5.0 -> 45.0 | Improved |
| apply Unified-GRPO to Qwen-2.5-VL-3B (ablation) | Qwen-2.5-VL-3B · MMT-Bench Document Understanding | accuracy: 70.0 -> 58.4 | Hurt |
| unfreeze visual encoder (ViT) during RL (ablation) | Unified-GRPO · reconstruction quality (qualitative) | image quality | Hurt |
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.
- OmniGen2 — OmniGen2: Towards Instruction-Aligned Multimodal Generation (CVPR 2026) · 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
- MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer (ICLR 2026 · similarity 0.90)
- Learning to Generate via Understanding: Understanding-Driven Intrinsic Rewarding for Unified Multimodal Models (CVPR 2026 · similarity 0.89)
- Reconstruction Alignment Improves Unified Multimodal Models (ICLR 2026 · similarity 0.89)
- UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm (CVPR 2026 · similarity 0.89)
- MARS: Mixture of Auto-Regressive Models for Fine-grained Text-to-image Synthesis (AAAI 2025 · similarity 0.89)
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models (NeurIPS 2025 · similarity 0.89)
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.
