We introduce a novel method for Photo Dating which estimates the year a photograph was taken by leveraging information from the faces of people present in the image. To facilitate this research, we publicly release CSFD-1.6M1, a new dataset containing over 1.6 million annotated faces, primarily from movie stills, with identity and birth year annotations. Uniquely, our dataset provides annotations for multiple individuals within a single image, enabling the study of multi-face information aggregation. We propose a probabilistic framework that formally combines visual evidence from modern face recognition and age estimation models, and career-based temporal priors to infer the photo capture year. Our experiments demonstrate that aggregating evidence from multiple faces consistently improves the performance and the approach significantly outperforms strong, scene-based baselines, particularly for images containing several identifiable individuals.

Comparing different age estimation methods poses a challenge due to the unreliability of published results stemming from inconsistencies in the benchmarking process. Previous studies have reported continuous performance improvements over the past decade using specialized methods; however, our findings challenge these claims. This paper identifies two trivial, yet persistent issues with the currently used evaluation protocol and describes how to resolve them. We offer an extensive comparative analysis for state-of-the-art facial age estimation methods. Surprisingly, we find that the performance differences between the methods are neg-ligible compared to the effect of other factors, such as facial alignment, facial coverage, image resolution, model architecture, or the amount of data used for pretraining. We use the gained insights to propose using FaRL as the back-bone model and demonstrate its effectiveness on all public datasets. We make the source code and exact data splits public on GitHub and in the supplementary material.

Binary statistical decision making involves choosing between two states based on statistical evidence. The optimal decision strategy is typically formulated through a constrained optimization problem, where both the objective and constraints are expressed as integrals involving two Lebesgue measurable functions, one of which represents the strategy being optimized. In this work, we present a comprehensive formulation of the binary decision making problem and provide a detailed characterization of the optimal solution. Our framework encompasses a wide range of well-known and recently proposed decision making problems as specific cases. We demonstrate how our generic approach can be used to derive the optimal decision strategies for these diverse instances. Our results offer a robust mathematical tool that simplifies the process of solving both existing and novel formulations of binary decision making problems which are in the core of many Machine Learning algorithms.

In classification with a reject option, the classifier is allowed in uncertain cases to abstain from prediction. The classical cost-based model of a reject option classifier requires the rejection cost to be defined explicitly. The alternative bounded-improvement model and the bounded-abstention model avoid the notion of the reject cost. The bounded-improvement model seeks a classifier with a guaranteed selective risk and maximal cover. The bounded-abstention model seeks a classifier with guaranteed cover and minimal selective risk. We prove that despite their different formulations the three rejection models lead to the same prediction strategy: the Bayes classifier endowed with a randomized Bayes selection function. We define the notion of a proper uncertainty score as a scalar summary of the prediction uncertainty sufficient to construct the randomized Bayes selection function. We propose two algorithms to learn the proper uncertainty score from examples for an arbitrary black-box classifier. We prove that both algorithms provide Fisher consistent estimates of the proper uncertainty score and demonstrate their efficiency in different prediction problems, including classification, ordinal regression, and structured output classification.

Structured output classifiers based on the framework of Markov Networks provide a transparent way to model statistical dependencies between output labels. The Markov Network (MN) classifier can be efficiently learned by the maximum margin method, which however requires expensive completely annotated examples. We extend the maximum margin algorithm for learning of unrestricted MN classifiers from examples with partially missing annotation of labels. The proposed algorithm translates learning into minimization of a novel loss function which is convex, has a clear connection with the supervised margin-rescaling loss, and can be efficiently optimized by first-order methods. We demonstrate the efficacy of the proposed algorithm on a challenging structured output classification problem where it beats deep neural network models trained from a much higher number of completely annotated examples, while the proposed method used only partial annotations.

We tackle the problem of dominant subject recognition (DSR), which aims at identifying the faces of the subject whose faces appear most frequently in a given collection of images. We propose a simple algorithm solving the DSR problem in a principled way via Bayesian learning. The proposed algorithm has complexity quadratic in the number of detected faces, and it provides labeling of images along with an accurate estimate of the prediction confidence. The prediction confidence permits using the algorithm in semiautomatic mode when only a subset of images with uncertain labels are corrected manually. We demonstrate on a challenging IJB-B database, that the algorithm significantly reduces the number of images that need to be manually annotated to get the perfect performance of face verification and face identification systems using the face database created by the method.

We propose a neural network which takes two inputs, a hair image and a face image, and produces an output image having the hair of the hair image seamlessly merged with the inner face of the face image. Our architecture consists of neural networks mapping the input images into a latent code of a pretrained StyleGAN2 which generates the output high-definition image. We propose an algorithm for training parameters of the architecture solely from synthetic images generated by the StyleGAN2 itself without the need of any annotations or external dataset of hairstyle images. We empirically demonstrate the effectiveness of our method in applications including hair-style transfer, hair generation for 3D morphable models, and hair-style interpolation. Fidelity of the generated images is verified by a user study and by a novel hairstyle metric proposed in the paper.

In classification with a reject option, the classifier is allowed in uncertain cases to abstain from prediction. The classical cost based model of an optimal classifier with a reject option requires the cost of rejection to be defined explicitly. An alternative bounded-improvement model, avoiding the notion of the reject cost, seeks for a classifier with a guaranteed selective risk and maximal cover. We prove that both models share the same class of optimal strategies, and we provide an explicit relation between the reject cost and the target risk being the parameters of the two models. An optimal rejection strategy for both models is based on thresholding the conditional risk defined by posterior probabilities which are usually unavailable. We propose a discriminative algorithm learning an uncertainty function which preserves ordering of the input space induced by the conditional risk, and hence can be used to construct optimal rejection strategies.

Supervised learning of convolutional neural networks (CNNs) for face recognition requires a large set of facial images each annotated with a single attribute label to be predicted. In this paper we propose a method for learning CNNs from weakly annotated images. The weak annotation in our setting means that a pair of an attribute label and a person identity label is assigned to a set of faces automatically detected in the image. The challenge is to link the annotation with the correct face. The weakly annotated images of this type can be collected by an automated process not requiring a human labor. We formulate learning from weakly annotated images as a maximum likelihood estimation of a parametric distribution describing the data. The ML problem is solved by an instance of EM algorithm which in its inner loop learns a CNN to perform given face recognition task. Experiments on age and gender estimation problem show that the proposed EM-CNN algorithm significantly outperforms the state-of-theart approach for dealing with this type of data.

Statistical learning theory relies on an assumption that the joint distributions of observations and labels are the same in training and testing data. However, this assumption is violated in many real world problems, such as training a detector of malicious network traffic that can change over time as a result of attacker's detection evasion efforts. We propose to address this problem by creating an optimized representation, which significantly increases the robustness of detectors or classifiers trained under this distributional shift. The representation is created from bags of samples (e.g. network traffic logs) and is designed to be invariant under shifting and scaling of the feature values extracted from the logs and under permutation and size changes of the bags. The invariance is achieved by combining feature histograms with feature self-similarity matrices computed for each bag and significantly reduces the difference between the training and testing data. The parameters of the representation, such as histogram bin boundaries, are learned jointly with the classifier. We show that the representation is effective for training a detector of malicious traffic, achieving 90% precision and 67% recall on samples of previously unseen malware variants.

A real-time algorithm for accurate localization of facial landmarks in a single monocular image is proposed. The algorithm is formulated as an optimization problem, in which the sum of responses of local classifiers is maximized with respect to the camera pose by fitting a generic (not a person-specific) 3D model. The algorithm simultaneously estimates a head position and orientation and detects the facial landmarks in the image. Despite being local, we show that the basin of attraction is large to the extent it can be initialized by a scanning window face detector. Other experiments on standard datasets demonstrate that the proposed algorithm outperforms a state-of-the-art landmark detector especially for non-frontal face images, and that it is capable of reliable and stable tracking for large set of viewing angles.

We propose a new stopping condition for a Support Vector Machine (SVM) solver which precisely reflects the objective of the Leave-One-Out error computation. The stopping condition guarantees that the output on an intermediate SVM solution is identical to the output of the optimal SVM solution with one data point excluded from the training set. A simple augmentation of a general SVM training algorithm allows one to use a stopping criterion equivalent to the proposed sufficient condition. A comprehensive experimental evaluation of our method shows consistent speedup of the exact LOO computation by our method, up to the factor of 13 for the linear kernel. The new algorithm can be seen as an example of constructive guidance of an optimization algorithm towards achieving the best attainable expected risk at optimal computational cost.

We have developed a new Linear Support Vector Machine (SVM) training algorithm called OCAS. Its computational effort scales linearly with the sample size. In an extensive empirical evaluation OCAS significantly outperforms current state of the art SVM solvers, like SVMlight, SVMperf and BMRM, achieving speedups of over 1,000 on some datasets over SVMlight and 20 over SVMperf, while obtaining the same precise Support Vector solution. OCAS even in the early optimization steps shows often faster convergence than the so far in this domain prevailing approximative methods SGD and Pegasos. Effectively parallelizing OCAS we were able to train on a dataset of size 15 million examples (itself about 32GB in size) in just 671 seconds --- a competing string kernel SVM required 97,484 seconds to train on 10 million examples sub-sampled from this dataset.

Evolutionary algorithms applied in real domain should profit from information about the local fitness function curvature. This paper presents an initial study of an evolutionary strategy with a novel approach for learning the covariance matrix of a Gaussian distribution. The learning method is based one stimation of the fitness landscape contour line between the selected and discarded individuals. The distribution learned this way is then used to generate new population members. The algorithm presented here is the first attempt to construct the Gaussian distribution this way and should beconsidered only a proof of concept; nevertheless, the empirical comparison on low-dimensional quadratic functions shows that our approach is viable and with respect to the number of evaluations needed to find a solution of certain quality, it is comparable to the state-of-the-art CMA-ES incase of sphere function and outperforms the CMA-ES in case of elliptical function.