Research map / Vision-language and multimodal / Human and embodied modeling
UniDDT: Unifying Multimodal Understanding and Generation with Decoupled Diffusion Transformer
Shuai Wang, Liang Li, Yang Chen, Ruopeng Gao, Yao Teng, Limin Wang
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 |
|---|---|---|---|
| add joint understanding-generation training after warmup (ablation) | UniDDT · GenEval (Native-UniDDT, by stage) | overall: 0.52 -> 0.60 | Improved |
| add duality-based post-training of the diffusion decoder (ablation) | UniDDT · GenEval (Native-UniDDT, by stage) | overall: 0.60 -> 0.72 | Improved |
| add 4o-style data fine-tuning (ablation) | UniDDT · GenEval (Native-UniDDT, by stage) | overall: 0.72 -> 0.87 | Improved |
| add joint training (ablation) | UniDDT · GenEval (Native-UniDDT, by stage) | counting: 0.32 -> 0.45 | Improved |
| add post-training (ablation) | UniDDT · GenEval (Native-UniDDT, by stage) | position: 0.41 -> 0.50 | Improved |
| add post-training (ablation) | UniDDT · GenEval (Native-UniDDT, by stage) | color attribution: 0.37 -> 0.61 | Improved |
| add understanding loss in joint training (vs w/o und) (ablation) | Latent-Native-UniDDT · GenEval | generation quality | Improved |
| add understanding loss in joint training (vs w/o und) (ablation) | Pixel-Native-UniDDT-B · GenEval | generation quality | Improved |
| use latent space instead of pixel space as unified visual space (ablation) | UniDDT · generation | generation quality | Improved |
| use pixel space instead of latent space (ablation) | VLM-UniDDT · understanding | MLLM benchmarks | No clear effect |
| use small time-shift instead of large time-shift in Noisy ViT warmup (ablation) | Noisy ViT encoder · understanding (OCR) | understanding | Improved |
| scale model from Native-UniDDT-L to XL | Native-UniDDT · GenEval | overall: 0.88 -> 0.89 | Improved |
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.
- Show-o2 7B — Show-o: One Single Transformer to Unify Multimodal Understanding and Generation (ICLR 2025) · better on 7 of 9 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
- Generation Enhances Understanding in Unified Multimodal Models via Multi-Representation Generation (ICML 2026 · similarity 0.92)
- Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation (CVPR 2025 · similarity 0.90)
- Show-o: One Single Transformer to Unify Multimodal Understanding and Generation (ICLR 2025 · similarity 0.90)
- Dual Diffusion for Unified Image Generation and Understanding (CVPR 2025 · similarity 0.90)
- DuoGen: Towards Autonomous Interleaved Multimodal Generation (CVPR 2026 · similarity 0.89)
- UniM: A Unified Any-to-Any Interleaved Multimodal Benchmark (CVPR 2026 · 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.
