Research map / Vision-language and multimodal / Human and embodied modeling
Mind-Omni: A Unified Multi-Task Framework for Brain-Vision-Language Modeling via Discrete Diffusion
Yizhuo Lu, Changde Du, Qingyu Shi, Hang Chen, Jie Peng, Liuyun Jiang, Shuangchen Zhao, Huiguang He
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 semantic alignment loss LSA to Brain Tokenizer (ablation) | Brain Tokenizer · fMRI-image/text retrieval (Top50) | B2I retrieval: 0.05 -> 0.58 | Improved |
| add semantic alignment loss LSA to Brain Tokenizer (ablation) | Brain Tokenizer · self-reconstruction | rPCC: 0.43 -> 0.64 | Improved |
| add perceptual alignment loss on top of LSA (ablation) | Brain Tokenizer · codebook usage | codebook usage: 62% -> 80% | Improved |
| enlarge codebook size or code dimension (ablation) | Brain Tokenizer · codebook usage | codebook usage: 70% (size 64, dim 16) -> 1% (dim 512) | Hurt |
| use progressive training instead of naive direct joint training (ablation) | Mind-Omni · B→I&T description and encoding | ROUGE: 27.31 -> 30.54 | Improved |
| initialize from pre-trained Muddit instead of training from scratch (ablation) | Mind-Omni · B→I&T description | BLEU1: 20.35 -> 29.12 | Improved |
| use Qwen2-VL enhanced captions instead of raw COCO captions (ablation) | Mind-Omni · B→I&T description | BLEU1: 24.37 -> 29.12 | Improved |
| reduce code dimension from 32 to 16 with more tokens (ablation) | Brain Tokenizer · encoding gPCC / decoding | gPCC: 0.126 -> 0.145 gPCC; BLEU1 29.56 -> 30.28 | Improved |
| jointly condition encoding on image and text (I&T→B) vs image only (ablation) | Mind-Omni · neural encoding voxel-level | gPCC: 0.157 -> 0.160 | Improved |
| jointly decode image and text (B→I&T) vs image only (B→I) (ablation) | Mind-Omni · NSD visual reconstruction | CLIP two-way accuracy: 66.7% -> 79.8% | Improved |
| jointly decode text and image (B→I&T) vs text only (B→T) (ablation) | Mind-Omni · Detail Description | BLEU1: 14.92 -> 29.12 | Improved |
| unified model vs specialized single-task models on B→I and B→T | Mind-Omni · holistic comparison | max-normalized score | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- UMBRAE — UMBRAE: Unified Multimodal Brain Decoding (ECCV 2024) · better on 0 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
- $i$MIND: Insightful Multi-subject Invariant Neural Decoding (NeurIPS 2025 · similarity 0.88)
- Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction (NeurIPS 2024 · similarity 0.88)
- Far from the Shallow: Brain-Predictive Reasoning Embedding through Residual Disentanglement (NeurIPS 2025 · similarity 0.87)
- Brain-OF: An Omnifunctional Foundation Model for fMRI, EEG and MEG (NeurIPS 2026 · similarity 0.87)
- MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding (NeurIPS 2026 · similarity 0.87)
- MindLLM: A Subject-Agnostic and Versatile Model for fMRI-to-text Decoding (ICML 2025 · 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.
