VoxelFormer: Parameter-Efficient Multi-Subject Visual Decoding from fMRI
On the map: Efficient LLM inference and serving → Visual token pruning in multimodal models
Abstract (arXiv)
Recent advances in fMRI-based visual decoding have enabled compelling reconstructions of perceived images. However, most approaches rely on subject-specific training, limiting scalability and practical deployment. We introduce VoxelFormer, a lightweight transformer architecture that enables multi-subject training for visual decoding from fMRI. VoxelFormer integrates a Token Merging Transformer (ToMer) for efficient voxel compression and a query-driven Q-Former that produces fixed-size neural representations aligned with the CLIP image embedding space. Evaluated on the 7T Natural Scenes Dataset, VoxelFormer achieves competitive retrieval performance on subjects included during training with significantly fewer parameters than existing methods. These results highlight token merging and query-based transformers as promising strategies for parameter-efficient neural decoding.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Visual token pruning in multimodal models, one of 12 directions in Efficient LLM inference and serving. Builds on 12 of its 16 references that are in the map. First posted on arXiv 10 Sep 2025.
Builds on
12 of 16 references.
References · 12
- 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
- 2014Microsoft COCO: Common Objects in Contextcited by 17,608
- 2022BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generationcited by 4,322
- 2020Implicit Neural Representations with Periodic Activation Functionscited by 2,238
- 2021Perceiver: General Perception with Iterative Attentioncited by 1,104
- 2021Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNetcited by 1,101
- 2022Token Merging: Your ViT But Fastercited by 1,021
- 2023Reconstructing the Mind's Eye: fMRI-to-Image with Contrastive Learning and Diffusion Priorscited by 160
- 2024MindEye2: Shared-Subject Models Enable fMRI-To-Image With 1 Hour of Datacited by 109
- 2023Natural scene reconstruction from fMRI signals using generative latent diffusioncited by 97
- 2024MindBridge: A Cross-Subject Brain Decoding Frameworkcited by 66
- 2017Machine learning for neural decodingcited by 45
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2024MindFormer: Semantic Alignment of Multi-Subject fMRI for Brain Decoding
- 2025NeuroSwift: A Lightweight Cross-Subject Framework for fMRI Visual Reconstruction of Complex Scenes
- 2022Semantic Brain Decoding: from fMRI to conceptually similar image reconstruction of visual stimuli
- 2024A multimodal LLM for the non-invasive decoding of spoken text from brain recordings
- 2023Contrast, Attend and Diffuse to Decode High-Resolution Images from Brain Activities
- 2025Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex
- 2024Optimized two-stage AI-based Neural Decoding for Enhanced Visual Stimulus Reconstruction from fMRI Data
- 2025Towards Interpretable Visual Decoding with Attention to Brain Representations