Pick the per-neuron JSON your admin sent you to start.
if you had reviewed this neuron earlier.
Welcome. This is a research demo of an auto-proofreading system developed by Fuming Yang and Sven Dorkenwald. Inference is performed end-to-end on a complete neuron from the MICrONS dataset in a single pass (~45 seconds). Ground-truth merge regions are pre-marked in a separate colour to aid comparison, but please note that this annotation is not exhaustive — regions outside the GT may still contain genuine merges, so please cross-reference the underlying EM imagery and the segmentation when making your judgement rather than relying on the GT colour alone.
Reviewers step through the suspect windows in the Decision panel and answer two questions per window:
For each suspect window, the system has already partitioned its tokens into one or more colored clusters (visible as soft ellipsoid blobs in the EM view). Your job: decide which of those clusters genuinely need to be split off as separate segments. Decisions persist in your browser's local storage and can be downloaded as JSON via Export decisions when you are finished. We greatly appreciate your contribution!
Keyboard: Y/N/S/U — Real merge?; 1–4 — toggle that-numbered cluster (in SPLIT WHICH); 0 — toggle Skip; →/Enter — next undecided; ← — step back one window; ↑/↓ or j/k — adjacent navigation.