Research map / Vision-language and multimodal / Human and embodied modeling
OmniMamba: Efficient and Unified Multimodal Understanding and Generation via State Space Models
Jialv Zou, Bencheng Liao, Qian Zhang, Wenyu Liu, Xinggang 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 |
|---|---|---|---|
| use decoupled (modality-specific) vocabularies instead of a unified vocabulary (ablation) | OmniMamba (Mamba-2-370M) · MS-COCO val | FID-30K: 19.1 -> 10.3 | Improved |
| use decoupled vocabularies instead of unified (ablation) | OmniMamba (Mamba-2-370M) · MME | score: 1036 -> 1100 | Improved |
| add task-specific LoRA (ablation) | OmniMamba (Mamba-2-370M) · MS-COCO val | FID-30K: 14.4 -> 10.3 | Improved |
| add task-specific LoRA (ablation) | OmniMamba (Mamba-2-370M) · POPE | accuracy: 81.2 -> 81.9 | Improved |
| add task-specific LoRA (ablation) | OmniMamba (Mamba-2-370M) · MME | score: 1003 -> 1100 | Improved |
| use OmniMamba instead of Show-o on understanding | OmniMamba · POPE | accuracy: 80.0 -> 86.3 | Improved |
| use OmniMamba instead of JanusFlow | OmniMamba · POPE / MME-P / VQAv2 | score: POPE 88.0 -> 86.3; MME-P 1333.1 -> 1290.6; GQA 60.3 -> 60.8 | Mixed |
| use OmniMamba instead of Show-o for generation | OmniMamba · MS-COCO val | FID-30K: 9.24 -> 5.50 | Improved |
| use Mamba-2 backbone instead of Transformer for image generation | OmniMamba · generation speed | images/s: 0.81 -> 5.68 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- Show-o — Show-o: One Single Transformer to Unify Multimodal Understanding and Generation (ICLR 2025) · better on 7 of 7 shared results
- JanusFlow — JanusFlow: Harmonizing Autoregression and Rectified Flow for Unified Multimodal Understanding and Generation (CVPR 2025) · better on 3 of 6 shared results
- Janus — Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation (CVPR 2025) · better on 3 of 5 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
- OpenING: A Comprehensive Benchmark for Judging Open-ended Interleaved Image-Text Generation (CVPR 2025 · similarity 0.90)
- MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer (ICLR 2026 · similarity 0.90)
- OmniBench: Towards The Future of Universal Omni-Language Models (NeurIPS 2025 · similarity 0.90)
- OmniGen2: Towards Instruction-Aligned Multimodal Generation (CVPR 2026 · similarity 0.90)
- MME-Unify: A Comprehensive Benchmark for Unified Multimodal Understanding and Generation Models (ICLR 2026 · similarity 0.89)
- UniGen: Enhanced Training & Test-Time Strategies for Unified Multimodal Understanding and Generation (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.
