Research map / Vision-language and multimodal / Human and embodied modeling
Modeling the Brain's Grammar: ROI-Guided fMRI Pretraining for Transferable and Interpretable Vision Decoding
Yulong Liu, Hua Xu, Yiyang Cai, Chunyang Jiang, Sirui Han, Yike Guo
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 MRL-inspired embedding compression (prefix masking of tokens) (ablation) | ROITok · NSD (40h, low-level only) | PixCorr: .426 -> .549 | Improved |
| add MRL-inspired embedding compression (ablation) | ROITok · NSD (40h) | Image retrieval: 97.3% -> 98.7% | Improved |
| two-stage finetuning (ROI embedders first, then all) instead of single-stage full finetuning (ablation) | ROITok 1h transfer · NSD (1h transfer) | Image retrieval: 66.1% -> 86.9% | Improved |
| add full finetuning stage after training only ROI embedders (ablation) | ROITok 1h transfer · NSD (1h transfer) | PixCorr: .337 -> .365 | Improved |
| use all ROI tokens instead of only the General token (ablation) | ROITok · NSD (1000-way retrieval, 40h) | retrieval accuracy: 96.2% -> 97.3% | Improved |
| combine V1-hV4 and Early tokens instead of Early alone (ablation) | ROITok · NSD (1000-way retrieval, 40h) | retrieval accuracy: 83.5% -> 87.9% | Improved |
| use IP-Adapter instead of diffusion prior for conditioning | ROITok · NSD (40h, full data) | SSIM: .351 -> .426 (SSIM); CLIP 95.2% -> 92.9% | Mixed |
| ROI-based tokenization instead of subject adapters (vs MindEye2) | cross-subject fMRI decoding · NSD (40h, full data) | PixCorr: .285 -> .475 | Improved |
| ROITok vs NeuroSwift on high-level metrics | cross-subject fMRI decoding · NSD (40h, full data) | CLIP: 97.1% (NeuroSwift) vs 95.2% | No clear effect |
| vary ROI mask ratio during pretraining (ablation) | ROITok · NSD (40h, subject 1) | retrieval and reconstruction | 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.
- MindTuner — MindTuner: Cross-Subject Visual Decoding with Visual Fingerprint and Semantic Correction (AAAI 2025) · better on 2 of 3 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
- See Through Their Minds: Learning Transferable Brain Decoding Models from Cross-Subject fMRI (AAAI 2025 · similarity 0.90)
- Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction (NeurIPS 2024 · similarity 0.90)
- MindEye2: Shared-Subject Models Enable fMRI-To-Image With 1 Hour of Data (ICML 2024 · similarity 0.89)
- MoRE-Brain: Routed Mixture of Experts for Interpretable and Generalizable Cross-Subject fMRI Visual Decoding (NeurIPS 2025 · similarity 0.88)
- Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex (NeurIPS 2025 · similarity 0.88)
- Multigranular Evaluation for Brain Visual Decoding (AAAI 2026 · 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.
