MEAD-3D

August 7, 2023 ยท View on GitHub

  • This dataset contains 3D reconstructions of the MEAD dataset, using the monocular 3D reconstruction algorithm in the paper 'Speech4Mesh: Speech-Assisted Monocular 3D Facial Reconstruction for Speech-Driven 3D Facial Animation' [ICCV 2023].

Comparison of the reconstruction results on MEAD (left) and VoxCeleb2 (right) datasets.

  • The dataset can be downloaded via BaiduNetDisk.

  • The dataset is organized as below:

    MEAD-3D

    Speaker ID

    video

    camera angle

    type of emotion

    intensity level

    video_id.npy

  • Each .npy file contains the following contents:

keysvaluesdescriptions
'verts'(N, 5023, 3)vertices reconstructed from each image
'beta'(43, 156)3DMM parameters (shape, expression, pose) of the FLAME reconstruction