AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

SEED: Towards More Accurate Semantic Evaluation for Visual Brain Decoding

Juhyeon Park, Peter Yongho Kim, Jiook Cha, Shinjae Yoo, Taesup Moon

ICLR 2026 · Vision-language and multimodal › Human and embodied modeling · approach: fMRI visual decoding

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 changedWhereMeasuredEffect
combine Object F1, Cap-Sim and EffNet into SEED rather than EffNet alone (ablation)SEED · NSD meta-evaluationKendall tau-b: .559 -> .621Improved
add Object F1 to EffNet (ablation)EffNet · NSD meta-evaluationKendall tau-b: .559 -> .602Improved
add Cap-Sim to EffNet (ablation)EffNet · NSD meta-evaluationKendall tau-b: .559 -> .583Improved
add the third metric (EffNet) to Object F1 + Cap-Sim (ablation)SEED · NSD meta-evaluationKendall tau-b: .566 -> .621Improved
replace fixed 82 categories with VLM-generated object categories (ablation)Object F1 · NSD meta-evaluationKendall tau-b: .516 -> .473Hurt
add object size weighting (ablation)Object F1 · NSD meta-evaluationKendall tau-b: .516 -> .517No clear effect
add object location weighting (ablation)Object F1 · NSD meta-evaluationKendall tau-b: .516 -> .517No clear effect
add object count partial credit (ablation)Object F1 · NSD meta-evaluationPearson: .708 -> .648Hurt
swap grounding model, captioner and text encoder (Yolo-World, BLIP-2, Qwen3-Embedding) (ablation)SEED · NSD meta-evaluationalignment with humanNo clear effect
evaluate SEED on a different dataset and decoder (GOD, Mind-Vis)SEED · GOD meta-evaluationKendall tau-b: .453 -> .477 vs EffNetImproved
use VQAScore-based similarity instead of Object F1 or Cap-Sim (ablation)semantic similarity metric · NSD meta-evaluationKendall tau-b: .434 (GIT VQAScore) vs .516 Object F1Hurt
use two-way identification metrics rather than correlation to EffNetexisting metrics · NSD meta-evaluationKendall tau-b: CLIP .368, Inception .330 vs EffNet .559Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy