Research map / Vision-language and multimodal / Multimodal LLMs
Exploring The Visual Feature Space for Multimodal Neural Decoding
Weihao Xia, Cengiz Oztireli
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 masked denoising loss to feature regression for brain alignment (ablation) | VINDEX · MG-BrainDUB descriptive captioning | CAPTURE: 0.4179 -> 0.4356 | Improved |
| make denoiser deeper (3 layers) or weight denoising loss at 2.0 (ablation) | VINDEX denoiser · MG-BrainDUB descriptive captioning | CAPTURE: 0.4356 -> 0.4096 | Hurt |
| align brain with 9-token nested CLIP features instead of single-encoder full features (ablation) | VINDEX · MG-BrainDUB descriptive captioning | CAPTURE: 0.4356 -> 0.5021 | Improved |
| nested feature with 144 tokens instead of 9 (ablation) | VINDEX · MG-BrainDUB descriptive captioning | CAPTURE: 0.5021 -> 0.1614 | Hurt |
| aggregate multi-layer SigLIP features (AF) as alignment target (ablation) | VINDEX · MG-BrainDUB descriptive captioning | CAPTURE: 0.4356 -> 0.0058 | Hurt |
| mix CLIP and DINO features (mixture of encoders) (ablation) | VINDEX · MG-BrainDUB descriptive captioning | CAPTURE: 0.4733 -> 0.3041 (SE vs ME, LLaVA-1.5 13B) | Hurt |
| use CLIP-336 instead of CLIP-224 target features (ablation) | VINDEX · MG-BrainDUB descriptive captioning | CAPTURE: 0.4877 -> 0.4120 (NF1) | Hurt |
| scale MLLM from LLaVA-1.5 7B to LLaVA-1.6 or 13B with single-encoder features (ablation) | VINDEX SE · MG-BrainDUB descriptive captioning | CAPTURE: 0.4356 -> 0.4733 (1.5 13B), 0.4684 (1.6 7B) | Mixed |
| use VINDEX instead of UMBRAE for brain grounding | VINDEX · BrainHub concept localization (All) | acc@0.5: 18.03 -> 18.10 | No clear effect |
| decode location directly from brain signals instead of from reconstructed images | VINDEX · BrainHub concept localization (All) | acc@0.5: 15.34 -> 18.10 | 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.
- UMBRAE — UMBRAE: Unified Multimodal Brain Decoding (ECCV 2024) · 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
- Multigranular Evaluation for Brain Visual Decoding (AAAI 2026 · similarity 0.91)
- Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction (NeurIPS 2024 · similarity 0.89)
- $i$MIND: Insightful Multi-subject Invariant Neural Decoding (NeurIPS 2025 · similarity 0.89)
- Breaking the Illusion: When Positive Meets Negative in Multimodal Decoding (CVPR 2026 · similarity 0.88)
- Finding Shared Decodable Concepts and their Negations in the Brain (ICLR 2025 · similarity 0.88)
- Revealing Vision-Language Integration in the Brain with Multimodal Networks (ICML 2024 · similarity 0.87)
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.
