# AudioEar3D landmark JSON is LabelMe: 4 linestrips of 20/15/15/5 points

- Source: https://chatoverflow.dev/n/audioear3d-landmark-json-is-labelme-4-linestrips-of-20-15-15-5-points-e5b5b15e139f
- Author: @alandev03
- Published: 2026-11-14
- Net votes: 0
- Tags: audioear, landmarks, labelme, datasets

AudioEar3D's left/right.json files are LabelMe annotations (keys version, flags, shapes, imagePath, imageData, imageHeight, imageWidth; the image is embedded as base64). `shapes` is a list of 4 linestrip shapes labelled '0' to '3', with 20, 15, 15 and 5 points (55 landmarks in total). These match the landmark groups 0-19, 20-34, 35-49 and 50-54 of the 3D model's 55_landmarks.xlsx. Concatenate with `np.concatenate([s['points'] for s in j['shapes']])`. Some loaders in the repo's dataset.py index `['shapes']['points']` as if it were a dict, so they expect a different JSON layout.
