With the rapid development of Deepfake synthesis technology, our information security and personal privacy have been severely threatened in recent years. To achieve a robust Deepfake detection, researchers attempt to exploit the joint spatial-temporal information in the videos, like using recurrent networks and 3D convolutional networks. However, these spatial-temporal models remain room to improve. Another general challenge for spatial-temporal models is that people do not clearly understand what these spatial-temporal models really learn. To address these two challenges, in this paper, we propose an Interpretable Spatial-Temporal Video Transformer (ISTVT), which consists of a novel decomposed spatial-temporal self-attention and a self-subtract mechanism to capture spatial artifacts and temporal inconsistency for robust Deepfake detection. Thanks to this decomposition, we propose to interpret ISTVT by visualizing the discriminative regions for both spatial and temporal dimensions via the relevance (the pixel-wise importance on the input) propagation algorithm. We conduct extensive experiments on large-scale datasets, including FaceForensics++, FaceShifter, DeeperForensics, Celeb-DF, and DFDC datasets. Our strong performance of intra-dataset and cross-dataset Deepfake detection demonstrates the effectiveness and robustness of our method, and our visualization-based interpretability offers people insights into our model.

Deep convolutional neural networks have recently proven extremely effective for difficult face recognition problems in uncontrolled settings. To train such networks, very large training sets are needed with millions of labeled images. For some applications, such as near-infrared (NIR) face recognition, such large training data sets are not publicly available and difficult to collect. In this paper, we propose a method to generate very large training data sets of synthetic images by compositing real face images in a given data set. We show that this method enables to learn models from as few as 10 000 training images, which perform on par with models trained from 500 000 images. Using our approach, we also obtain state-of-the-art results on the CASIA NIR-VIS2.0 heterogeneous face recognition data set.

MixUp is an effective data augmentation method to regularize deep neural networks via random linear interpolations between pairs of samples and their labels. It plays an important role in model regularization, semisupervised learning (SSL), and domain adaption. However, despite its empirical success, its deficiency of randomly mixing samples has poorly been studied. Since deep networks are capable of memorizing the entire data set, the corrupted samples generated by vanilla MixUp with a badly chosen interpolation policy will degrade the performance of networks. To overcome overfitting to corrupted samples, inspired by metalearning (learning to learn), we propose a novel technique of learning to a mixup in this work, namely, MetaMixUp. Unlike the vanilla MixUp that samples interpolation policy from a predefined distribution, this article introduces a metalearning-based online optimization approach to dynamically learn the interpolation policy in a data-adaptive way (learning to learn better). The validation set performance via metalearning captures the noisy degree, which provides optimal directions for interpolation policy learning. Furthermore, we adapt our method for pseudolabel-based SSL along with a refined pseudolabeling strategy. In our experiments, our method achieves better performance than vanilla MixUp and its variants under SL configuration. In particular, extensive experiments show that our MetaMixUp adapted SSL greatly outperforms MixUp and many state-of-the-art methods on CIFAR-10 and SVHN benchmarks under the SSL configuration.

Facial pose variation is one of the major factors making face recognition (FR) a challenging task. One popular solution is to convert non-frontal faces to frontal ones on which FR is performed. Rotating faces causes facial pixel value changes. Therefore, existing CNN-based methods learn to synthesize frontal faces in color space. However, this learning problem in a color space is highly non-linear, causing the synthetic frontal faces to lose fine facial textures. In this paper, we take the view that the nonfrontal-frontal pixel changes are essentially caused by geometric transformations (rotation, translation, and so on) in space. Therefore, we aim to learn the nonfrontal-frontal facial conversion in the spatial domain rather than the color domain to ease the learning task. To this end, we propose an appearance-flow-based face frontalization convolutional neural network (A3F-CNN). Specifically, A3F-CNN learns to establish the dense correspondence between the non-frontal and frontal faces. Once the correspondence is built, frontal faces are synthesized by explicitly "moving" pixels from the non-frontal one. In this way, the synthetic frontal faces can preserve fine facial textures. To improve the convergence of training, an appearance-flow-guided learning strategy is proposed. In addition, generative adversarial network loss is applied to achieve a more photorealistic face, and a face mirroring method is introduced to handle the self-occlusion problem. Extensive experiments are conducted on face synthesis and pose invariant FR. Results show that our method can synthesize more photorealistic faces than the existing methods in both the controlled and uncontrolled lighting environments. Moreover, we achieve a very competitive FR performance on the Multi-PIE, LFW and IJB-A databases.