Tracks & data.

EEG epoch · 32 ch Alljoined-1.6M

rank

Held-out gallery

DINOv2 embedding space

Sealed participants and images are unseen Top-5 acc ↑

Visual decoding is most useful when it recognizes novel content rather than a fixed catalogue learned during training. Given a single EEG epoch recorded during natural-image viewing, identify the viewed image by ranking held-out candidates in a frozen DINOv2 embedding space. The sealed final uses participants and images absent from the public development data, testing cross-subject and cross-stimulus transfer. Warm-up isolates the unseen-image shift on THINGS-EEG2. Research-grade and consumer-grade datasets add source and hardware variability while keeping the task anchored to time-locked visual responses.Metric: Top-5 retrieval accuracy against the held-out candidate set, with higher scores indicating better performance.Warm-up evaluation: THINGS-EEG2 test set · Sealed evaluation: private 2026 Alljoined cohort · Start kit and baseline: THINGS-EEG2

Decode one of three cued mental commands from EEG: kinesthetic motor imagery, mental subtraction, or word generation. The 2026 cohort follows 20 participants across six sessions using the Graz and BrainHero interfaces. Public training data contain early sessions from ten participants. Sealed evaluation combines later sessions from those participants with recordings from entirely new participants. This tests robustness to session drift and inter-person variability without access to hidden labels.Metric: Balanced accuracy averaged across participant, session, and recording-context cells, with higher scores indicating better performance.Current warm-up and baseline: public PROTEUS three-command task · Sealed: later sessions and new participantsDataset: PROTEUS BCI Bordeaux is public on NEMAR and used by the Codabench warm-up.

Predict from continuous four-channel home EEG how many seconds remain until the first N2 epoch. During Muse evaluation, five-second windows arrive one at a time in chronological order. Models may use past context within the current recording, and their state is reset before the next recording. Evaluation combines new nights from sleepers represented in training with recordings from entirely unseen sleepers, testing night-to-night and inter-person generalization under motion artifacts, impedance changes, and occasional channel dropout.Metric: Weighted binned mean absolute error (W-bMAE) in seconds, with greater weight on errors closer to sleep onset. Seen-sleeper and unseen-sleeper W-bMAE are macro-averaged for the final ranking. Lower scores indicate better performance.Current warm-up and baseline: public Muse causal-stream task · Warm-up score: per-recording W-bMAE mean · Sealed: private Muse cohortDataset: Muse Sleep-Onset EEG is public on NEMAR and used by the Codabench warm-up.

Translating wrist muscle activity into hand motion is a core challenge for wearable control systems that must work beyond a calibrated laboratory setup. Predict trajectories of 20 hand-joint angles from 16-channel wrist sEMG recorded during everyday movement. Evaluation includes users and movement stages never seen during training, plus unseen combinations of known users and stages. This tests robustness to anatomy, wristband placement, and kinematic context, and asks whether a daily wearable interface can transfer to a new user without person-specific retraining.Metric: Models return joint angles in radians. Codabench reports their mean absolute angular error in degrees, with lower scores indicating better performance.Warm-up evaluation: EMG2Pose user-stage test set · Sealed evaluation: private 2026 Meta Reality Labs cohort · Start kit and baseline: EMG2Pose