Airborne laser scanning (ALS) point cloud data are suitable for digital terrain model (DTM) extraction given its high accuracy in elevation. Existing filtering algorithms that eliminate non-ground points mostly depend on terrain feature assumptions or representations; these assumptions result in errors when the scene is complex. This paper proposes a new method for ground point extraction based on deep learning using deep convolutional neural networks (CNN). For every point with spatial context, the neighboring points within a window are extracted and transformed into an image. Then, the classification of a point can be treated as the classification of an image; the point-to-image transformation is carefully crafted by considering the height information in the neighborhood area. After being trained on approximately 17 million labeled ALS points, the deep CNN model can learn how a human operator recognizes a point as a ground point or not. The model performs better than typical existing algorithms in terms of error rate, indicating the significant potential of deep-learning-based methods in feature extraction from a point cloud.

3D Morphable Model (3DMM) based methods have achieved great success in recovering 3D face shapes from single-view images. However, the facial textures recovered by such methods lack the fidelity as exhibited in the input images. Recent works demonstrate high-quality facial texture recovering with generative networks trained from a large-scale database of high-resolution UV maps of face textures, which is hard to prepare and not publicly available. In this paper, we introduce a method to reconstruct 3D facial shapes with high-fidelity textures from single-view images in the wild, without the need to capture a large-scale face texture database. The main idea is to refine the initial texture generated by a 3DMM based method with facial details from the input image. To this end, we propose to use graph convolutional networks to reconstruct the detailed colors for the mesh vertices instead of reconstructing the UV map. Experiments show that our method can generate high-quality results and outperforms state-of-the-art methods in both qualitative and quantitative comparisons.

Automatic cloud extraction from satellite imagery is an important task for many applications in remote sensing. Humans can easily identify various clouds from satellite images based on the visual features of cloud. In this study, a method of automatic cloud detection is proposed based on object classification of image features. An image is first segmented into superpixels so that the descriptor of each superpixel can be computed to form a feature vector for classification. The support vector machine algorithm is then applied to discriminate cloud and noncloud regions. Thereafter, the GrabCut algorithm is used to extract more accurate cloud regions. The key of the method is to deal with the highly varying patterns of clouds. The bag-of-words (BOW) model is used to construct the compact feature vectors from densely extracted local features, such as dense scale-invariant feature transform (SIFT). The algorithm is tested using 101 RapidEye and 86 Landsat images with many cloud patterns. These images achieve 89.2% of precision, 87.8% of recall for RapidEye, 85.8% of precision, and 83.9% of recall for Landsat. The experiments show that the method is insensitive to the number of codewords in the codebook construction of the BOW.

Head pose estimation from RGB images without depth information is a challenging task due to the loss of spatial information as well as large head pose variations in the wild. The performance of existing landmark-free methods remains unsatisfactory as the quality of estimated pose is inferior. In this paper, we propose a novel three-branch network architecture, termed as Feature Decoupling Network (FDN), a more powerful architecture for landmark-free head pose estimation from a single RGB image. In FDN, we first propose a feature decoupling (FD) module to explicitly learn the discriminative features for each pose angle by adaptively recalibrating its channel-wise responses. Besides, we introduce a cross-category center (CCC) loss to constrain the distribution of the latent variable subspaces and thus we can obtain more compact and distinct subspaces. Extensive experiments on both in-the-wild and controlled environment datasets demonstrate that the proposed method outperforms other state-of-the-art methods based on a single RGB image and behaves on par with approaches based on multimodal input resources.

We present a deep-learning-based method to automatically compute skin weights for skeleton-based deformation of production characters. Given a character mesh and its associated skeleton hierarchy in rest pose, our method constructs a graph for the mesh, each node of which encodes the mesh-skeleton attributes of a vertex. An end-to-end deep graph convolution network is then introduced to learn the mesh-skeleton binding patterns from a set of character models with skin weights painted by artists. The network can be used to predict the skin weight map for a new character model, which describes how the skeleton hierarchy influences the mesh vertices during deformation. Our method is designed to work for non-manifold meshes with multiple disjoint or intersected components, which are common in game production and require complex skeleton hierarchies for animation control. We tested our method on the datasets of two commercial games. Experiments show that the predicted skin weight maps can be readily applied to characters in the production pipeline to generate high-quality deformations.