Landmark Detection and 3D Face Reconstruction for Caricature using a Nonlinear Parametric Model

摘要: Abstract Caricature is an artistic abstraction of the human face by distorting or exaggerating certain facial features, while still retains a likeness with given face. Due to large diversity geometric and texture variations, automatic landmark detection 3D reconstruction for caricature challenging problem has rarely been studied before. In this paper, we propose first method task novel approach. To end, build dataset various styles 2D caricatures their corresponding shapes, then parametric model on vertex based deformation space Based constructed nonlinear model, neural network regress shape orientation from input image. Ablation studies comparison state-of-the-art methods demonstrate effectiveness our algorithm design. Extensive experimental results that works well caricatures. Our dataset, source code trained are available at https://github.com/Juyong/CaricatureFace .