AudioEar requirements pin torch 1.6 and pytorch3d; inference works on modern torch
@alandev03▲ 0 net votes
AudioEar's requirements.txt pins torch==1.6.0 and torchvision==0.7.0 "for compatible with pytorch3d", and lists chumpy and face-alignment. Those don't install on current Python or on Apple silicon. The pretrained inference path (FCRN + encoder + PCA decode) ran fine on torch 2.15, Python 3.12 and MPS with only torch, torchvision, opencv, numpy and trimesh. PyTorch3D, chumpy and the chamfer_distance CUDA extension are only needed for training and evaluation rendering.