Recently, deep learning based 3D face reconstruction methods have shown promising results in both quality and efficiency. However, training deep neural networks typically requires a large volume of data, whereas face images with ground-truth 3D face shapes are scarce. In this paper, we propose a novel deep 3D face reconstruction approach that 1) leverages a robust, hybrid loss function for weakly-supervised learning which takes into account both low-level and perception-level information for supervision, and 2) performs multi-image face reconstruction by exploiting complementary information from different images for shape aggregation. Our method is fast, accurate, and robust to occlusion and large pose. We provide comprehensive experiments on three datasets, systematically comparing our method with fifteen recent methods and demonstrating its state-of-the-art performance.

This paper presents a Neural Aggregation Network (NAN) for video face recognition. The network takes a face video or face image set of a person with a variable number of face images as its input, and produces a compact, fixed-dimension feature representation for recognition. The whole network is composed of two modules. The feature embedding module is a deep Convolutional Neural Network (CNN) which maps each face image to a feature vector. The aggregation module consists of two attention blocks which adaptively aggregate the feature vectors to form a single feature inside the convex hull spanned by them. Due to the attention mechanism, the aggregation is invariant to the image order. Our NAN is trained with a standard classification or verification loss without any extra supervision signal, and we found that it automatically learns to advocate high-quality face images while repelling low-quality ones such as blurred, occluded and improperly exposed faces. The experiments on IJB-A, YouTube Face, Celebrity-1000 video face recognition benchmarks show that it consistently outperforms naive aggregation methods and achieves the state-of-the-art accuracy.

No abstract

No abstract

We propose a framework based on Generative Adversarial Networks to disentangle the identity and attributes of faces, such that we can conveniently recombine different identities and attributes for identity preserving face synthesis in open domains. Previous identity preserving face synthesis processes are largely confined to synthesizing faces with known identities that are already in the training dataset. To synthesize a face with identity outside the training dataset, our framework requires one input image of that subject to produce an identity vector, and any other input face image to extract an attribute vector capturing, e.g., pose, emotion, illumination, and even the background. We then recombine the identity vector and the attribute vector to synthesize a new face of the subject with the extracted attribute. Our proposed framework does not need to annotate the attributes of faces in any way. It is trained with an asymmetric loss function to better preserve the identity and stabilize the training process. It can also effectively leverage large amounts of unlabeled training face images to further improve the fidelity of the synthesized faces for subjects that are not presented in the labeled training face dataset. Our experiments demonstrate the efficacy of the proposed framework. We also present its usage in a much broader set of applications including face frontalization, face attribute morphing, and face adversarial example detection.

[1] A UV multilevel partitioning method (UV-MLP) is developed to solve scalar wave three-dimensional (3-D) scattering problem. The method consists of setting up a table of transmitting and receiving block size and their separation using fast coarse-coarse sampling. For a specific scattering problem with given geometry, the scattering structure is partitioned into multilevel blocks. By looking up the rank in the static problem, the impedance matrix for a given transmitting and receiving block is expressed into a product of U and V matrix. In this paper the method is illustrated by applying to a 3-D scattering problem of random nonpenetrable rough surface. The cases of Dirichelt and Neumann boundary conditions are treated. Numerical simulation results are illustrated. For 65,536 boundary unknowns on a rough surface, and using a single processor of 2.66 GHz, it takes about 34 CPU min and 1.8 Gb of memory to compute the solution using conjugate gradient iterations and multilevel UV to accelerate the matrix-column vector multiplication.INDEX TERMS: 0644 Electromagnetics: Numerical methods; 0659 Electromagnetics: Random media and rough surfaces; 0669 Electromagnetics: Scattering and diffraction; KEYWORDS: wave scattering, fast solver, rough surface Citation: Tsang, L., Q. Li, P. Xu, D. Chen, and V. Jandhyala (2004), Wave scattering with UV multilevel partitioning method: 2. Three-dimensional problem of nonpenetrable surface scattering, Radio Sci., 39, RS5011,

Abstract:Object-based change detection (OBCD) has recently been receiving increasing attention as a result of rapid improvements in the resolution of remote sensing data. However, some OBCD issues relating to the segmentation of high-resolution images remain to be explored. For example, segmentation units derived using different segmentation strategies, segmentation scales, feature space, and change detection methods have rarely been assessed. In this study, we have tested four common unsupervised change detection methods using different segmentation strategies and a series of segmentation scale parameters on two WorldView-2 images of urban areas. We have also evaluated the effect of adding extra textural and Normalized Difference Vegetation Index (NDVI) information instead of using only spectral information. Our results indicated that change detection methods performed better at a medium scale than at a fine scale where close to the pixel size. Multivariate Alteration Detection (MAD) always outperformed the other methods tested, at the same confidence level. The overall accuracy appeared to benefit from using a two-date segmentation strategy rather than single-date segmentation. Adding textural and NDVI information appeared to reduce detection accuracy, but the magnitude of this reduction was not consistent across the different unsupervised methods and segmentation strategies. We conclude that a two-date segmentation strategy is useful for change detection in high-resolution imagery, but that the optimization of thresholds is critical for unsupervised change detection methods. Advanced methods need be explored that can take advantage of additional textural or other parameters.

No abstract