Facial morphs created between two identities resemble both of the faces used to create the morph. Consequently, humans and machines are prone to mistake morphs made from two identities for either of the faces used to create the morph. This vulnerability has been exploited in "morph attacks" in security scenarios. Here, we asked whether the "other-race effect" (ORE)-the human advantage for identifying own- vs. other-race faces-exacerbates morph attack susceptibility for humans. We also asked whether face-identification performance in a deep convolutional neural network (DCNN) is affected by the race of morphed faces. Caucasian (CA) and East-Asian (EA) participants performed a face-identity matching task on pairs of CA and EA face images in two conditions. In the morph condition, different-identity pairs consisted of an image of identity "A" and a 50/50 morph between images of identity "A" and "B". In the baseline condition, morphs of different identities never appeared. As expected, morphs were identified mistakenly more often than original face images. Of primary interest, morph identification was substantially worse for cross-race faces than for own-race faces. Similar to humans, the DCNN performed more accurately for original face images than for morphed image pairs. Notably, the deep network proved substantially more accurate than humans in both cases. The results point to the possibility that DCNNs might be useful for improving face identification accuracy when morphed faces are presented. They also indicate the significance of the race of a face in morph attack susceptibility in applied settings.

Common and important applications of person identification occur at distances and viewpoints in which the face is not visible or is not sufficiently resolved to be useful. We examine body shape as a biometric across distance and viewpoint variation. We propose an approach that combines standard object classification networks with representations based on linguistic (word-based) descriptions of bodies. Algorithms with and without linguistic training were compared on their ability to identify people from body shape in images captured across a large range of distances/views (close-range, 100m, 200m, 270m, 300m, 370m, 400m, 490m, 500m, 600m, and at elevated pitch in images taken by an unmanned aerial vehicle [UAV]). Accuracy, as measured by identity-match ranking and false accept errors in an open-set test, was surprisingly good. For identity-ranking, linguistic models were more accurate for close-range images, whereas non-linguistic models fared better at intermediary distances. Fusion of the linguistic and non-linguistic embeddings improved performance at all but the farthest distance. Although the non-linguistic model yielded fewer false accepts at most distances, fusion of the linguistic and non-linguistic models decreased false accepts for all but the UAV images. We conclude that linguistic and non-linguistic representations of body shape can offer complementary identity information for bodies that can improve identification in applications of interest.

Deep convolutional neural networks (DCNNs) have achieved human-level accuracy in face identification (Phillips et al., 2018), though it is unclear how accurately they discriminate highly-similar faces. Here, humans and a DCNN performed a challenging face-identity matching task that included identical twins. Participants ( N = 87) viewed pairs of face images of three types: same-identity, general imposters (different identities from similar demographic groups), and twin imposters (identical twin siblings). The task was to determine whether the pairs showed the same person or different people. Identity comparisons were tested in three viewpoint-disparity conditions: frontal to frontal, frontal to 45° profile, and frontal to 90° profile. Accuracy for discriminating matched-identity pairs from twin-imposter pairs and general-imposter pairs was assessed in each viewpoint-disparity condition. Humans were more accurate for general-imposter pairs than twin-imposter pairs, and accuracy declined with increased viewpoint disparity between the images in a pair. A DCNN trained for face identification (Ranjan et al., 2018) was tested on the same image pairs presented to humans. Machine performance mirrored the pattern of human accuracy, but with performance at or above all humans in all but one condition. Human and machine similarity scores were compared across all image-pair types. This item-level analysis showed that human and machine similarity ratings correlated significantly in six of nine image-pair types [range r = 0.38 to r = 0.63], suggesting general accord between the perception of face similarity by humans and the DCNN. These findings also contribute to our understanding of DCNN performance for discriminating high-resemblance faces, demonstrate that the DCNN performs at a level at or above humans, and suggest a degree of parity between the features used by humans and the DCNN.

We present SfSNet, an end-to-end learning framework for producing an accurate decomposition of an unconstrained human face image into shape, reflectance and illuminance. SfSNet is designed to reflect a physical lambertian rendering model. SfSNet learns from a mixture of labeled synthetic and unlabeled real-world images. This allows the network to capture low-frequency variations from synthetic and high-frequency details from real images through the photometric reconstruction loss. SfSNet consists of a new decomposition architecture with residual blocks that learns a complete separation of albedo and normal. This is used along with the original image to predict lighting. SfSNet produces significantly better quantitative and qualitative results than state-of-the-art methods for inverse rendering and independent normal and illumination estimation. We also introduce a companion network, SfSMesh, that utilizes normals estimated by SfSNet to reconstruct a 3D face mesh. We demonstrate that SfSMesh produces face meshes with greater accuracy than state-of-the-art methods on real-world images.

In recent years, visible-spectrum face verification systems have been shown to match the performance of experienced forensic examiners. However, such systems are ineffective in low-light and nighttime conditions. Thermal face imagery, which captures body heat emissions, effectively augments the visible spectrum, capturing discriminative facial features in scenes with limited illumination. Due to the increased cost and difficulty of obtaining diverse, paired thermal and visible spectrum datasets, not many algorithms and large-scale benchmarks for low-light recognition are available. This paper presents an algorithm that achieves state-of-the-art performance on both the ARL-VTF and TUFTS multi-spectral face datasets. Importantly, we study the impact of face alignment, pixel-level correspondence, and identity classification with label smoothing for multi-spectral face synthesis and verification. We show that our proposed method is widely applicable, robust, and highly effective. In addition, we show that the proposed method significantly outperforms face frontalization methods on profile-to-frontal verification. Finally, we present MILAB-VTF(B), a challenging multi-spectral face dataset that is composed of paired thermal and visible videos. To the best of our knowledge, with face data from 400 subjects, this dataset represents the most extensive collection of publicly available indoor and long-range outdoor thermal-visible face imagery. Lastly, we show that our end-to-end thermal-to-visible face verification system provides strong performance on the MILAB-VTF(B) dataset.

Face recognition networks encode information about sensitive attributes while being trained for identity classification. Such encoding has two major issues: (a) it makes the face representations susceptible to privacy leakage (b) it appears to contribute to bias in face recognition. However, existing bias mitigation approaches generally require end-to-end training and are unable to achieve high verification accuracy. Therefore, we present a descriptor-based adversarial de-biasing approach called ‘Protected Attribute Suppression System (PASS)’. PASS can be trained on top of descriptors obtained from any previously trained high-performing network to classify identities and simultaneously reduce encoding of sensitive attributes. This eliminates the need for end-to-end training. As a component of PASS, we present a novel discriminator training strategy that discourages a network from encoding protected attribute information. We show the efficacy of PASS to reduce gender and skintone information in descriptors from SOTA face recognition networks like Arcface. As a result, PASS descriptors outperform existing baselines in reducing gender and skintone bias on the IJB-C dataset, while maintaining a high verification accuracy.

Face de-identification (or “masking”) algorithms have been developed in response to the prevalent use of video recordings in public places. We evaluated the success of face identity masking for human perceivers and a deep convolutional neural network (DCNN). Eight de-identification algorithms were applied to videos of drivers’ faces, while they actively operated a motor vehicle. These masks were pre-selected to be applicable to low-quality video and to maintain coarse information about facial actions. Humans studied high-resolution images to learn driver identities and were tested on their recognition of active drivers in low-resolution videos. Faces in the videos were either unmasked or were masked by one of the eight algorithms. When participants were tested immediately after learning (Experiment 1), all masks reduced identification, with six of eight masks reducing identification to extremely poor performance. In a second experiment, two of the most effective masks were tested after a delay of 7 or 28 days. The delay did not further reduce identification of the masked faces. In all masked conditions, participants maintained stringent decision criteria, with low confidence in recognition, further indicating the effectiveness of the masks. Next, the DCNN performed an identity-matching task between high-resolution images and masked videos—a task analogous to that done by humans. The pattern of accuracy for the DCNN mirrored some, but not all, aspects of human performance, highlighting the need to test the effectiveness of identity masking for both humans and machines. The DCNN was also tested on its ability to match identity between masked and unmasked versions of the same video, based only on the face. DCNN performance for the eight masks offers insight into the nature of the information in faces that is coded in these networks.

As deep networks become increasingly accurate at recognizing faces, it is vital to understand how these networks process faces. While these networks are solely trained to recognize identities, they also contain face related information such as sex, age, and pose of the face even when the networks are not trained to learn these attributes. We introduce expressivity as a measure of how much a feature vector informs us about an attribute, where a feature vector can be from internal or final layers of a network. Expressivity is computed by a second neural network whose inputs are features and attributes. The output of the second neural network approximates the mutual information between feature vectors and an attribute. We investigate the expressivity for two different deep convolutional neural network (DCNN) architectures: a Resnet-101 and an Inception Resnet v2. In the final fully connected layer of the networks, we found the order of expressivity for facial attributes to be Age > Sex > Yaw. Additionally, we studied the changes in the encoding of facial attributes over training iterations. We found that as training progresses, expressivities of yaw, sex, and age decrease. Our technique can be a tool for investigating the sources of bias in a network and a step towards explaining the network's identity decisions.

Unconstrained video-based face recognition is a challenging problem due to significant within-video variations caused by pose, occlusion and blur. To tackle this problem, an effective idea is to propagate the identity from high-quality faces to low-quality ones through contextual connections, which are constructed based on context such as body appearance. However, previous methods have often propagated erroneous information due to lack of uncertainty modeling of the noisy contextual connections. In this paper, we propose the Uncertainty-Gated Graph (UGG), which conducts graph-based identity propagation between tracklets, which are represented by nodes in a graph. UGG explicitly models the uncertainty of the contextual connections by adaptively updating the weights of the edge gates according to the identity distributions of the nodes during inference. UGG is a generic graphical model that can be applied at only inference time or with end-to-end training. We demonstrate the effectiveness of UGG with state-of-the-art results in the recently released challenging Cast Search in Movies and IARPA Janus Surveillance Video Benchmark dataset.

For a given identity in a face dataset, there are certain iconic images which are more representative of the subject than others. In this paper, we explore the problem of computing the iconicity of a face. The premise of the proposed approach is as follows: For an identity containing a mixture of iconic and non iconic images, if a given face cannot be successfully matched with any other face of the same identity, then the iconicity of the face image is low. Using this information, we train a Siamese Multi-Layer Perceptron network, such that each of its twins predict iconicity scores of the image feature pair, fed in as input. We observe the variation of the obtained scores with respect to covariates such as blur, yaw, pitch, roll and occlusion to demonstrate that they effectively predict the quality of the image and compare it with other existing metrics. Furthermore, we use these scores to weight features for template-based face verification and compare it with media averaging of features.

Existing approaches for spatio-temporal action detection in videos are limited by the spatial extent and temporal duration of the actions. In this paper, we present a modular system for spatio-temporal action detection in untrimmed surveillance videos. We propose a two stage approach. The first stage generates dense spatio-temporal proposals using hierarchical clustering and temporal jittering techniques on frame-wise object detections. The second stage is a Temporal Refinement I3D (TRI-3D) network that performs action classification and temporal refinement on the generated proposals. The object detection-based proposal generation step helps in detecting actions occurring in a small spatial region of a video frame, while temporal jittering and refinement helps in detecting actions of variable lengths. Experimental results on an unconstrained surveillance action detection dataset - DIVA - show the effectiveness of our system. For comparison, the performance of our system is also evaluated on the THUMOS'14 temporal action detection dataset.

Attributes are human describable features, which have been used successfully for face, object, and activity recognition. Facial attributes are intuitive descriptions of faces and have proven to be very useful in face recognition and verification. Despite their usefulness, to date there is only one large-scale facial attribute dataset, CelebA. Impressive results have been achieved on this dataset, but it exhibits a variety of very significant biases. As CelebA contains mostly frontal idealized images of celebrities, it is difficult to generalize a model trained on this data for use on another dataset (of non celebrities). A typical approach to dealing with imbalanced data involves sampling the data in order to balance the positive and negative labels, however, with a multi-label problem this becomes a non-trivial task. By sampling to balance one label, we affect the distribution of other labels in the data. To address this problem, we introduce a novel Selective Learning method for deep networks which adaptively balances the data in each batch according to the desired distribution for each label. The bias in CelebA can be corrected for in this way, allowing the network to learn a more robust attribute model. We argue that without this multi-label balancing, the network cannot learn to accurately predict attributes that are poorly represented in CelebA. We demonstrate the effectiveness of our method on the problem of facial attribute prediction on CelebA, LFWA, and the new University of Maryland Attribute Evaluation Dataset (UMD-AED), outperforming the state-of-the-art on each dataset.

In this paper, we consider the problem of grouping a collection of unconstrained face images in which the number of subjects is not known. We propose an unsupervised clustering algorithm called Deep Density Clustering (DDC) which is based on measuring density affinities between local neighborhoods in the feature space. By learning the minimal covering sphere for each neighborhood, information about the underlying structure is encapsulated. The encapsulation is also capable of locating high-density region of the neighborhood, which aids in measuring the neighborhood similarity. We theoretically show that the encapsulation asymptotically converges to a Parzen window density estimator. Our experiments show that DDC is a superior candidate for clustering unconstrained faces when the number of subjects is unknown. Unlike conventional linkage and density-based methods that are sensitive to the selection operating points, DDC attains more consistent and improved performance. Furthermore, the density-aware property reduces the difficulty in finding appropriate operating points.

Domain Adaptation is an actively researched problem in Computer Vision. In this work, we propose an approach that leverages unsupervised data to bring the source and target distributions closer in a learned joint feature space. We accomplish this by inducing a symbiotic relationship between the learned embedding and a generative adversarial network. This is in contrast to methods which use the adversarial framework for realistic data generation and retraining deep models with such data. We demonstrate the strength and generality of our approach by performing experiments on three different tasks with varying levels of difficulty: (1) Digit classification (MNIST, SVHN and USPS datasets) (2) Object recognition using OFFICE dataset and (3) Domain adaptation from synthetic to real data. Our method achieves state-of-the art performance in most experimental settings and by far the only GAN-based method that has been shown to work well across different datasets such as OFFICE and DIGITS.

We present SfSNet, an end-to-end learning framework for producing an accurate decomposition of an unconstrained human face image into shape, reflectance and illuminance. SfSNet is designed to reflect a physical lambertian rendering model. SfSNet learns from a mixture of labeled synthetic and unlabeled real world images. This allows the network to capture low frequency variations from synthetic and high frequency details from real images through the photometric reconstruction loss. SfSNet consists of a new decomposition architecture with residual blocks that learns a complete separation of albedo and normal. This is used along with the original image to predict lighting. SfSNet produces significantly better quantitative and qualitative results than state-of-the-art methods for inverse rendering and independent normal and illumination estimation.

Recent research progress in facial attribute recognition has been dominated by small improvements on the only large-scale publicly available benchmark dataset, CelebA [18]. We propose to extend attribute prediction research to unconstrained videos. Applying attribute models trained on CelebA - a still image dataset - to video data highlights several major problems with current models, including the lack of consideration for both time and motion. Many facial attributes (e.g. gender, hair color) should be consistent throughout a video, however, current models do not produce consistent results. We introduce two methods to increase the consistency and accuracy of attribute responses in videos: a temporal coherence constraint, and a motionattention mechanism. Both methods work on weakly labeled data, requiring attribute labels for only one frame in a sequence, which we call the anchor frame. The temporal coherence constraint moves the network responses of non-anchor frames toward the responses of anchor frames for each sequence, resulting in more stable and accurate attribute predictions. We use the motion between anchor and non-anchor video frames as an attention mechanism, discarding the information from parts of the non-anchor frame where no motion occurred. This motion-attention focuses the network on the moving parts of the non-anchor frames (i.e. the face). Since there is no large-scale video dataset labeled with attributes, it is essential for attribute models to be able to learn from weakly labeled data. We demonstrate the effectiveness of the proposed methods by evaluating them on the challenging YouTube Faces video dataset [31]. The proposed motion-attention and temporal coherence methods outperform attribute models trained on CelebA, as well as those fine-tuned on video data. To the best of our knowledge, this paper is the first to address the problem of facial attribute prediction in video.

In this paper, we present a new video-based face identification algorithm, where the target (i.e., person of interest) in the probe video is only annotated once with a face bounding box in a frame and the video may consist of multiple shots. Most video face identification techniques assume that the video is of single shot, and thus the bounding boxes of the target face can be extracted by tracking a face across the video frames. Nevertheless, such automatic annotation is vulnerable to the drifting of the face tracker, and the face tracking algorithm is inadequate to associate the face images of the target across multiple shots. In this paper, we propose a target face association (TFA) technique that retrieves a set of representative face images in a given video that are likely to have the same identity as the target face. These face images are then utilized to construct a robust face representation of the target face for searching the corresponding subject in the gallery. Since two faces that appear in the same video frame cannot belong to the same person, such cannot-link constraints are utilized for learning a target-specific linear classifier for establishing the intra/inter-shot face association of the target. Experimental results on the newly released JANUS challenge set 3 (JANUS CS3) dataset show that our method generates robust representations from target-annotated videos and demonstrates good performance for the task of video-based face identification problem.

Face recognition algorithms based on deep convolutional neural networks (DCNNs) have made progress on the task of recognizing faces in unconstrained viewing conditions. These networks operate with compact feature-based face representations derived from learning a very large number of face images. Although the learned feature sets produced by DCNNs can be highly robust to changes in viewpoint, illumination, and appearance, little is known about the nature of the face code that emerges at the top level of these networks. We analyzed the DCNN features produced by two recent face recognition algorithms. In the first set of experiments, we used the top-level features from the DCNNs as input into linear classifiers aimed at predicting metadata about the images. The results showed that the DCNN features contained surprisingly accurate information about the yaw and pitch of a face, and about whether the input face came from a still image or a video frame. In the second set of experiments, we measured the extent to which individual DCNN features operated in a view-dependent or view-invariant manner for different identities. We found that view-dependent coding was a characteristic of the identities rather than the DCNN features– with some identities coded consistently in a view-dependent way and others in a view-independent way. In our third analysis, we visualized the DCNN feature space for 24,000+ images of 500 identities. Images in the center of the space were uniformly of low quality (e.g., extreme views, face occlusion, poor contrast, low resolution). Image quality increased monotonically as a function of distance from the origin. This result suggests that image quality information is available in the DCNN features, such that consistently average feature values reflect coding failures that reliably indicate poor or unusable images. Combined, the results offer insight into the coding mechanisms that support robust representation of faces in DCNNs.

We present a multi-purpose algorithm for simultaneous face detection, face alignment, pose estimation, gender recognition, smile detection, age estimation and face recognition using a single deep convolutional neural network (CNN). Theproposed method employs a multi-task learning framework that regularizes the shared parameters of CNN and builds a synergy among different domains and tasks. Extensive experiments show that the network has a better understanding of face and achieves state-of-the-art result for most of these tasks.

Style transfer is an important task in which the style of a source image is mapped onto that of a target image. The method is useful for synthesizing derivative works of a particular artist or specific painting. This work considers targeted style transfer, in which the style of a template image is used to alter only part of a target image. For example, an artist may wish to alter the style of only one particular object in a target image without altering the object's general morphology or surroundings. This is useful, for example, in augmented reality applications (such as the recently released Pokémon go), where one wants to alter the appearance of a single real-world object in an image frame to make it appear as a cartoon. Most notably, the rendering of real-world objects into cartoon characters has been used in a number of films and television show, such as the upcoming series Son of Zorn. We present a method for targeted style transfer that simultaneously segments and stylizes single objects selected by the user. The method uses a Markov random field model to smooth and anti-alias outlier pixels near object boundaries, so that stylized objects naturally blend into their surroundings.

Recent progress in face detection (including keypoint detection), and recognition is mainly being driven by (i) deeper convolutional neural network architectures, and (ii) larger datasets. However, most of the large datasets are maintained by private companies and are not publicly available. The academic computer vision community needs larger and more varied datasets to make further progress. In this paper, we introduce a new face dataset, called UMDFaces, which has 367,888 annotated faces of 8,277 subjects. We also introduce a new face recognition evaluation protocol which will help advance the state-of-the-art in this area. We discuss how a large dataset can be collected and annotated using human annotators and deep networks. We provide human curated bounding boxes for faces. We also provide estimated pose (roll, pitch and yaw), locations of twenty-one key-points and gender information generated by a pre-trained neural network. In addition, the quality of keypoint annotations has been verified by humans for about 115,000 images. Finally, we compare the quality of the dataset with other publicly available face datasets at similar scales.

Although deep learning has yielded impressive performance for face recognition, many studies have shown that different networks learn different feature maps: while some networks are more receptive to pose and illumination others appear to capture more local information. Thus, in this work, we propose a deep heterogeneous feature fusion network to exploit the complementary information present in features generated by different deep convolutional neural networks (DCNNs) for template-based face recognition, where a template refers to a set of still face images or video frames from different sources which introduces more blur, pose, illumination and other variations than traditional face datasets. The proposed approach efficiently fuses the discriminative information of different deep features by 1) jointly learning the non-linear high-dimensional projection of the deep features and 2) generating a more discriminative template representation which preserves the inherent geometry of the deep features in the feature space. Experimental results on the IARPA Janus Challenge Set 3 (Janus CS3) dataset demonstrate that the proposed method can effectively improve the recognition performance. In addition, we also present a series of covariate experiments on the face verification task for in-depth qualitative evaluations for the proposed approach.

We have collected a new face data set that will facilitate research in the problem of frontal to profile face verification `in the wild'. The aim of this data set is to isolate the factor of pose variation in terms of extreme poses like profile, where many features are occluded, along with other `in the wild' variations. We call this data set the Celebrities in Frontal-Profile (CFP) data set. We find that human performance on Frontal-Profile verification in this data set is only slightly worse (94.57% accuracy) than that on Frontal-Frontal verification (96.24% accuracy). However we evaluated many state-of-the-art algorithms, including Fisher Vector, Sub-SML and a Deep learning algorithm. We observe that all of them degrade more than 10% from Frontal-Frontal to Frontal-Profile verification. The Deep learning implementation, which performs comparable to humans on Frontal-Frontal, performs significantly worse (84.91% accuracy) on Frontal-Profile. This suggests that there is a gap between human performance and automatic face recognition methods for large pose variation in unconstrained images.

2-D face recognition in the presence of large pose variations presents a significant challenge. When comparing a frontal image of a face to a near profile image, one must cope with large occlusions, non-linear correspondences, and significant changes in appearance due to viewpoint. Stereo matching has been used to handle these problems, but performance of this approach degrades with large pose changes. We show that some of this difficulty is due to the effect that foreshortening of slanted surfaces has on window-based matching methods, which are needed to provide robustness to lighting change. We address this problem by designing a new, dynamic programming stereo algorithm that accounts for surface slant. We show that on the CMU PIE dataset this method results in significant improvements in recognition performance.

Face recognition across pose is a problem of fundamental importance in computer vision. We propose to address this problem by using stereo matching to judge the similarity of two, 2D images of faces seen from different poses. Stereo matching allows for arbitrary, physically valid, continuous correspondences. We show that the stereo matching cost provides a very robust measure of similarity of faces that is insensitive to pose variations. To enable this, we show that, for conditions common in face recognition, the epipolar geometry of face images can be computed using either four or three feature points. We also provide a straightforward adaptation of a stereo matching algorithm to compute the similarity between faces. The proposed approach has been tested on the CMU PIE data set and demonstrates superior performance compared to existing methods in the presence of pose variation. It also shows robustness to lighting variation.

We propose using stereo matching for 2-D face recognition across pose. We match one 2-D query image to one 2-D gallery image without performing 3-D reconstruction. Then the cost of this matching is used to evaluate the similarity of the two images. We show that this cost is robust to pose variations. To illustrate this idea we built a face recognition system on top of a dynamic programming stereo matching algorithm. The method works well even when the epipolar lines we use do not exactly fit the viewpoints. We have tested our approach on the PIE dataset. In all the experiments, our method demonstrates effective performance compared with other algorithms.

In non-deterministic and dynamic environments, such as the RoboCup simulation league, it is necessary to simplify the search space to manage action selection in real time. In this work, we present Chimps, a team for RoboCup simulation league that uses an accuracy-based evolutionary reinforcement learning mechanism, called XCS to achieve this simplification. XCS is a Genetic Classifier System, with generalization capacities; we use them for the evolution of individual behavior's rules. We modified an existing team, 11Monkeys, that used static rules for individual action selection, adding an XCS to learn in real time over the outcome of individual actions. We found that our extension enhanced the team's performance.