This is a work report repository about Virtual Digital Human, including 3D Human Head, 3D Human Bodies, 3D Human Motions, 3D Human Textures, and 3D Virtual Tryon.
I have compiled the research results into the work report in PDF format, PDF can be found here.
This repository mainly contains one demos, as well as their implementation details and pipelines
(The other files in git are unoptimized effects that can be referenced)
- The following image :
(a).FFHQ-UV method: The high fidelity and low fidelity human faces of the original FFHQ-UV algorithm
(b).Face detail method: similarity is more than 90%, basically identical, and has strong adaptability
(c).Comparative learning method: 80% similarity, able to handle skin tone and lighting issues, only able to have a similar appearance
(a).Using Facial-Detail-Methods to reconstruct 某人's 3D face,
(b).Using contrastive learning methods to reconstruct 鲁迅's 3D face，
and all Add mouth shaped voice broadcasting and actions to generate 3D digital humans.
 |
 |
Different replacement methods were used to replace the heads of 27554smpl and smplx with flames.
27554Smpl is the body model of Demo1(3D Virtual Tryon), which contains 27554 vertices. Smplx is the body model of Demo2(3D Human Reconstruction).
The two replacement methods are as follows :
We have referred to many open source projects, and we have listed a few of them.
SMPL
SMPLX
textured_smplx
Mesh
Flame
FLame2SMPLX
FACIAL
FACEGOOD-Audio2Face
voca
meshtalk
FaceFormer
Open3D
hifi3dface
VideoPose3D
RingNet
DECA
PIXIE
ICON
neural-blend-shapes
deep-motion-editing
RigNet
MultiGarmentNetwork
TailorNet
pix2surf
vto-garment-collisions
vto-learning-based-animation
VirtualBones
BCNet
squeezeformer
TensorFlowASR
TensorFlowTTS
Meta-TTS
FastSpeech2