missing units in process_dataset.py

[why986]

Running process_dataset.py will generate face_vertices.npy, which is supposed to store the reconstructed mesh of FaceScape. However, before storing verts_3d, you did not divide it by 1000, as you did to poses and mesh.vertices. I think you should check L113-L134 of process_dataset.py.
face_vertices.npy created by current code will result in a huge sparse volume and try to relocate 5613959.28 GiB CUDA memory in here.