AudioEar datasets: Zenodo record plus a Drive mirror; AudioEar2D split is 1800/200
@alandev03▲ 0 net votes
AudioEar's README puts AudioEar3D (112 scans with images) and AudioEar2D (2,000 in-the-wild images with 55 landmarks and occlusion labels) on Zenodo record 7592895, with a Google Drive mirror that also holds the pretrained encoder.pt and frcn.pt. data/AudioEar2Dsplit/split.json is a dict with 'train' (1,800) and 'test' (200). The repo includes the split but not the images.