Papers
  • Initial page
  • SIGGRAPH 2017
    • Decomposing Images into Layers via RGB-space Geometry
    • Image-based Reconstruction of Wire Art
    • Bounding Proxies for Shape Approximation
    • GRASS: Generative Recursive Autoencoders for Shape Structures
    • DeepSketch2Face: A Deep Learning Based Sketching System for 3D Face and Caricature Modeling
    • Real-Time User-Guided Image Colorization with Learned Deep Priors
    • Spectral Remapping for Image Downscaling
  • CVPR 2017
    • Learning Shape Abstractions by Assembling Volumetric Primitives
    • Synthesizing 3D Shapes via Modeling Multi-View Depth Maps and Silhouettes with Deep Generative Netwo
    • Deep Image Matting
    • Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields
    • DeMoN: Depth and Motion Network for Learning Monocular Stereo
    • A Unified Approach of Multi-scale Deep and Hand-crafted Features for Defocus Estimation
    • One-Shot Video Object Segmentation
  • SIGGRAPH ASIA 2017
    • Deep Reverse Tone Mapping
    • HDR image reconstruction from a single exposure using deep CNNs
    • BigSUR: Large-scale Structured Urban Reconstruction
    • Adaptive Synthesis of Indoor Scenes via Activity-Associated Object Relation Graphs
    • Robust Structure Simplification for Hex Re-meshing
    • Seamless: Seam erasure and seam-aware decoupling of shape from mesh resolution
    • ComplementMe: Weakly-Supervised Component Suggestions for 3D Modeling
Powered by GitBook
On this page
  • Abstract
  • 问题描述
  • 主要贡献
  • Main Effect/Pipeline
  • 效果
  • 使用流程
  • 网络结构
  • Discussion
  • 缺点
  • Results
  1. SIGGRAPH 2017

DeepSketch2Face: A Deep Learning Based Sketching System for 3D Face and Caricature Modeling

http://i.cs.hku.hk/~xghan/Projects/ske2face.htm

PreviousGRASS: Generative Recursive Autoencoders for Shape StructuresNextReal-Time User-Guided Image Colorization with Learned Deep Priors

Last updated 6 years ago

Keyword: Face Modeling, Face Database, Deep Learning, Face Caricatures, Gestures, Sketch-Based Modeling

Abstract

问题描述

  • 基于深度学习的，由草图构建夸张面部表情三维模型的算法

主要贡献

  • 草图到三维面部模型的算法实现，提供给用户交互修改的功能

  • 基于CNN且对草图形状特征敏感的网络

  • 提供了大量不同夸张程度的表情数据（训练集）

Main Effect/Pipeline

效果

使用流程

网络结构

Discussion

缺点

  • 当输入的草图过于夸张，或者不符合训练集假定的输入要求，会出现奇怪的结果

  • 基于形变的三维模型难以表现细腻的褶皱等特征

  • CNN网络可以替换为更先进的U-Nets/GAN等提高精度

Results