Adventures in Kedreamix' Digital World

REAL3D-PORTRAIT: ONE-SHOT REALISTIC 3D TALKING PORTRAIT SYNTHESISPaper : https://arxiv.org/pdf/2401.08503.pdf Project : https://real3dportrait.github.io/ Code : https://github.com/yerfor/Real3DPortrait Rebuttal: https://real3dportrait.github.io/static/pages/rebuttal.html 摘要 (1) 研究背景：说话人像生成旨在根据驱动条件（动作序列或驱动音频）合成说话人像视频。这是一个计算机图形学和计算机视觉中长期存在的跨模态任务，具有视频会议和虚拟现实 (VR) 等多项实际应用。先前的 2D 方法可以产生逼真的视频，这要归功于生成对抗网络 (GAN) 的强大功能。然而，由于缺乏显式的 3D 建模，这些 2D 方法在头部大幅移动时会面临变形伪影和不真实的失真。在过去的几年中，基于神经辐射场 (NeRF) 的 3D ...

NeRF 方向最新论文已更新，请持续关注 Update in 2024-03-13 SMURF Continuous Dynamics for Motion-Deblurring Radiance Fields

3DGS 方向最新论文已更新，请持续关注 Update in 2024-03-13 StyleGaussian Instant 3D Style Transfer with Gaussian Splatting

Talking Head Generation 方向最新论文已更新，请持续关注 Update in 2024-03-13 A Comparative Study of Perceptual Quality Metrics for Audio-driven Talking Head Videos

Diffusion Models 方向最新论文已更新，请持续关注 Update in 2024-03-13 Bridging Different Language Models and Generative Vision Models for Text-to-Image Generation

Diffusion Models 方向最新论文已更新，请持续关注 Update in 2024-03-11 VideoElevator Elevating Video Generation Quality with Versatile Text-to-Image Diffusion Models

Blendshapes泛指3D定点动画的制作方式 (Maya里面称之为 blend shapes ，而3DS Max里称之为morph targets) ，在3D动画中用的比较多，尤其是人脸动画的制作，通过blendshape来驱动角色的面部表情。

NeRF 方向最新论文已更新，请持续关注 Update in 2024-03-09 DART Implicit Doppler Tomography for Radar Novel View Synthesis

3DGS 方向最新论文已更新，请持续关注 Update in 2024-03-09 3DGStream On-the-Fly Training of 3D Gaussians for Efficient Streaming of Photo-Realistic Free-Viewpoint Videos

Talking Head Generation 方向最新论文已更新，请持续关注 Update in 2024-03-09 FaceChain-ImagineID Freely Crafting High-Fidelity Diverse Talking Faces from Disentangled Audio