Research map / Vision-language and multimodal / Cross-modal learning
HBridge: H-Shape Bridging of Heterogeneous Experts for Unified Multimodal Understanding and Generation
Xiang Wang, Zhifei Zhang, He Zhang, Zhe Lin, Yuqian Zhou, Qing Liu, Shiwei Zhang, Yijun Li, Shaoteng Liu, Haitian Zheng, Jason Kuen, Yuehuan Wang and 2 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 |
|---|---|---|---|
| bridge only middle layers (skip first and last 6) instead of all layers (ablation) | HBridge 0.5B+4B · DPG-Bench | overall: 78.19 -> 80.03 | Improved |
| bridge only middle layers (M=N=6) (ablation) | HBridge 0.5B+4B · GenEval | overall: 0.61 -> 0.66 | Improved |
| skip too many layers (M=N=10) (ablation) | HBridge 0.5B+4B · DPG-Bench | overall: 80.03 -> 69.05 | Hurt |
| skip 8 layers per end instead of 6 (ablation) | HBridge 0.5B+4B · GenEval | overall: 0.66 -> 0.54 | Hurt |
| add semantic reconstruction tokens (ablation) | HBridge 0.5B+4B · DPG-Bench | overall: 79.57 -> 80.03 | Improved |
| use deep (mid-layer) fusion instead of shallow last-layer fusion (ablation) | HBridge 0.5B+4B · DPG-Bench | overall: 74.53 -> 80.03 | Improved |
| initialize generative expert from pretrained diffusion model instead of VLM-initialized DiT (ablation) | HBridge · image quality at 40k steps | visual quality | Improved |
| use larger understanding expert (7B vs 0.5B) (ablation) | HBridge · image quality | visual quality | Improved |
| add mid-layer semantic bridge to BAGEL (ablation) | BAGEL 7B+7B · GenEval | overall: 0.82 -> 0.84 | Improved |
| add mid-layer bridge and semantic tokens to BAGEL (ablation) | BAGEL 7B+7B · ImgEdit-Bench | overall: 3.20 -> 3.38 | Improved |
| finetune BAGEL on collected data only (ablation) | BAGEL 7B+7B · DPG-Bench | overall: 85.07 -> 85.20 | No clear effect |
| asymmetric 7B+4B HBridge versus symmetric BAGEL 7B+7B | HBridge · DPG-Bench | overall: 85.07 -> 85.23 | 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.
- OmniGen2 — OmniGen2: Towards Instruction-Aligned Multimodal Generation (CVPR 2026) · better on 4 of 4 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
- Rosetta: Composable Native Multimodal Pretraining (NeurIPS 2026 · similarity 0.88)
- DuoGen: Towards Autonomous Interleaved Multimodal Generation (CVPR 2026 · similarity 0.87)
- Generation Enhances Understanding in Unified Multimodal Models via Multi-Representation Generation (ICML 2026 · similarity 0.87)
- ROVER: Benchmarking Reciprocal Cross-Modal Reasoning for Omnimodal Generation (ICLR 2026 · similarity 0.87)
- Breaking Dual Bottlenecks: Evolving Unified Multimodal Models into Self-Adaptive Interleaved Visual Reasoners (ICML 2026 · similarity 0.87)
- LaTtE-Flow: Layerwise Timestep-Expert Flow-based Transformer (ICML 2026 · similarity 0.86)
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.
