Research map / Vision-language and multimodal / Human and embodied modeling
BrainJanus: A Unified Model for Understanding and Generation across Brain, Vision, and Language
Haitao Wu, Qirui Zhang, Zhouheng Yao, Shangquan Sun, Qihao Zheng, Mianxin Liu, Chi Zhang, Wanli Ouyang, Chunfeng Song, Changqing Zhang, Jiamin Wu
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 |
|---|---|---|---|
| train the model on brain-to-image pairs instead of only brain-to-text pairs (ablation) | BrainJanus · NSD brain-to-image reconstruction | CLIP: 77.3% -> 94.4% | Improved |
| train the model on brain-to-image pairs instead of only brain-to-text pairs (ablation) | BrainJanus · NSD brain-to-image reconstruction | PixCorr: 0.077 -> 0.173 | Improved |
| use detailed Qwen captions as ground truth instead of short COCO captions | BrainJanus · NSD brain-to-caption | BERT: 42.12 -> 38.12 | Mixed |
| use BrainJanus instead of MindLLM for brain-to-caption | brain-to-caption · NSD brain-to-caption (GT = Qwen detailed caption) | CIDEr: 2.11 -> 41.32 | Improved |
| use BrainJanus instead of Transformer Encoding for brain encoding | brain encoding · NSD brain encoding (image branch) | Inception: 70.4% -> 72.8% | Improved |
| use autoregressive BrainJanus instead of diffusion-based MindEye2 for brain-to-image | brain-to-image · NSD brain-to-image reconstruction | PixCorr: 0.322 -> 0.173 | Hurt |
| use BrainJanus instead of MindEye2 for brain-to-image | brain-to-image · NSD brain-to-image reconstruction | CLIP: 93.0% -> 94.4% | Improved |
| increase codebook size beyond 128 or 256 in the brain tokenizer (ablation) | brain tokenizer · tokenizer ablation | reconstruction and semantic metrics | No clear effect |
| zero-pad ground-truth visual embedding as fake fMRI (Padding Hacking) | visual-to-fMRI evaluation protocol · visual-to-fMRI synthesis | CLIP: 94.3% -> 100.0% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- UMBRAE — UMBRAE: Unified Multimodal Brain Decoding (ECCV 2024) · better on 7 of 7 shared results
- MindLLM — MindLLM: A Subject-Agnostic and Versatile Model for fMRI-to-text Decoding (ICML 2025) · better on 6 of 6 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
- Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction (NeurIPS 2024 · similarity 0.90)
- MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding (NeurIPS 2026 · similarity 0.88)
- BrainSCUBA: Fine-Grained Natural Language Captions of Visual Cortex Selectivity (ICLR 2024 · similarity 0.88)
- Brain Harmony: A Multimodal Foundation Model Unifying Morphology and Function into 1D Tokens (NeurIPS 2025 · similarity 0.88)
- $i$MIND: Insightful Multi-subject Invariant Neural Decoding (NeurIPS 2025 · similarity 0.88)
- CSBrain: A Cross-scale Spatiotemporal Brain Foundation Model for EEG Decoding (NeurIPS 2025 · similarity 0.88)
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.
