This repository contains two parts of the NeuroMM-2026 multimodal seizure detection challenge dataset:
Train + Val (root annotations/ + archives/): 25,426 EEG samples (20,298 train / 5,128 val) with labels and matching pre-extracted visual features from 7 vision backbones.
Test-phase candidate set (candidate/): the test-phase inputs only — no labels. Predictions are submitted to and scored on the official Codabench leaderboards;… See the full description on the dataset page:
https://huggingface.co/datasets/NeuroMM/NeuroMM-2026.
