We introduce a self-supervised representation learning method based on the task of temporal alignment between videos. The method trains a network using temporal cycle consistency (TCC), a differentiable cycle-consistency loss that can be used to find correspondences across time in multiple videos. The resulting per-frame embeddings can be used to align videos by simply matching frames using the nearest-neighbors in the learned embedding space. To evaluate the power of the embeddings, we densely label the Pouring and Penn Action video datasets for action phases. We show that (i) the learned embeddings enable few-shot classification of these action phases, significantly reducing the supervised training requirements; and (ii) TCC is complementary to other methods of self-supervised learning in videos, such as Shuffle and Learn and Time-Contrastive Networks. The embeddings are also used for a number of applications based on alignment (dense temporal correspondence) between video pairs, including transfer of metadata of synchronized modalities between videos (sounds, temporal semantic labels), synchronized playback of multiple videos, and anomaly detection. Project webpage: https://sites.google.com/view/temporal-cycle-consistency .
Navigation is a rich and well-grounded problem domain that drives progress in many different areas of research: perception, planning, memory, exploration, and optimisation in particular. Historically these challenges have been separately considered and solutions built that rely on stationary datasets - for example, recorded trajectories through an environment. These datasets cannot be used for decision-making and reinforcement learning, however, and in general the perspective of navigation as an interactive learning task, where the actions and behaviours of a learning agent are learned simultaneously with the perception and planning, is relatively unsupported. Thus, existing navigation benchmarks generally rely on static datasets (Geiger et al., 2013; Kendall et al., 2015) or simulators (Beattie et al., 2016; Shah et al., 2018). To support and validate research in end-to-end navigation, we present StreetLearn: an interactive, first-person, partially-observed visual environment that uses Google Street View for its photographic content and broad coverage, and give performance baselines for a challenging goal-driven navigation task. The environment code, baseline agent code, and the dataset are available at http://streetlearn.cc
The objective of this paper is speaker recognition "in the wild"-where utterances may be of variable length and also contain irrelevant signals. Crucial elements in the design of deep networks for this task are the type of trunk (frame level) network, and the method of temporal aggregation. We propose a powerful speaker recognition deep network, using a "thin-ResNet" trunk architecture, and a dictionary-based NetVLAD or GhostVLAD layer to aggregate features across time, that can be trained end-to-end. We show that our network achieves state of the art performance by a significant margin on the VoxCeleb1 test set for speaker recognition, whilst requiring fewer parameters than previous methods. We also investigate the effect of utterance length on performance, and conclude that for "in the wild" data, a longer length is beneficial.
We introduce the Action Transformer model for recognizing and localizing human actions in video clips. We repurpose a Transformer-style architecture to aggregate features from the spatiotemporal context around the person whose actions we are trying to classify. We show that by using high-resolution, person-specific, class-agnostic queries, the model spontaneously learns to track individual people and to pick up on semantic context from the actions of others. Additionally its attention mechanism learns to emphasize hands and faces, which are often crucial to discriminate an action - all without explicit supervision other than boxes and class labels. We train and test our Action Transformer network on the Atomic Visual Actions (AVA) dataset, outperforming the state-of-the-art by a significant margin - more than 7.5% absolute (40% relative) improvement, using only raw RGB frames as input.
True video understanding requires making sense of non-lambertian scenes where the color of light arriving at the camera sensor encodes information about not just the last object it collided with, but about multiple mediums -- colored windows, dirty mirrors, smoke or rain. Layered video representations have the potential of accurately modelling realistic scenes but have so far required stringent assumptions on motion, lighting and shape. Here we propose a learning-based approach for multi-layered video representation: we introduce novel uncertainty-capturing 3D convolutional architectures and train them to separate blended videos. We show that these models then generalize to single videos, where they exhibit interesting abilities: color constancy, factoring out shadows and separating reflections. We present quantitative and qualitative results on real world videos.
Learning a deep neural network requires solving a challenging optimization problem: it is a high-dimensional, non-convex and non-smooth minimization problem with a large number of terms. The current practice in neural network optimization is to rely on the stochastic gradient descent (SGD) algorithm or its adaptive variants. However, SGD requires a hand-designed schedule for the learning rate. In addition, its adaptive variants tend to produce solutions that generalize less well on unseen data than SGD with a hand-designed schedule. We present an optimization method that offers empirically the best of both worlds: our algorithm yields good generalization performance while requiring only one hyper-parameter. Our approach is based on a composite proximal framework, which exploits the compositional nature of deep neural networks and can leverage powerful convex optimization algorithms by design. Specifically, we employ the Frank-Wolfe (FW) algorithm for SVM, which computes an optimal step-size in closed-form at each time-step. We further show that the descent direction is given by a simple backward pass in the network, yielding the same computational cost per iteration as SGD. We present experiments on the CIFAR and SNLI data sets, where we demonstrate the significant superiority of our method over Adam, Adagrad, as well as the recently proposed BPGrad and AMSGrad. Furthermore, we compare our algorithm to SGD with a hand-designed learning rate schedule, and show that it provides similar generalization while converging faster. The code is publicly available at https://github.com/oval-group/dfw.
With the propensity for deep learning models to learn unintended signals from data sets there is always the possibility that the network can `cheat' in order to solve a task. In the instance of data sets for visual kinship verification, one such unintended signal could be that the faces are cropped from the same photograph, since faces from the same photograph are more likely to be from the same family. In this paper we investigate the influence of this artefactual data inference in published data sets for kinship verification. To this end, we obtain a large dataset, and train a CNN classifier to determine if two faces are from the same photograph or not. Using this classifier alone as a naive classifier of kinship, we demonstrate near state of the art results on five public benchmark data sets for kinship verification - achieving over 90% accuracy on one of them. Thus, we conclude that faces derived from the same photograph are a strong inadvertent signal in all the data sets we examined, and it is likely that the fraction of kinship explained by existing kinship models is small.
Nearly all existing counting methods are designed for a specific object class. Our work, however, aims to create a counting model able to count any class of object. To achieve this goal, we formulate counting as a matching problem, enabling us to exploit the image self-similarity property that naturally exists in object counting problems. We make the following three contributions: first, a Generic Matching Network (GMN) architecture that can potentially count any object in a class-agnostic manner; second, by reformulating the counting problem as one of matching objects, we can take advantage of the abundance of video data labeled for tracking, which contains natural repetitions suitable for training a counting model. Such data enables us to train the GMN. Third, to customize the GMN to different user requirements, an adapter module is used to specialize the model with minimal effort, i.e. using a few labeled examples, and adapting only a small fraction of the trained parameters. This is a form of few-shot learning, which is practical for domains where labels are limited due to requiring expert knowledge (e.g. microbiology). We demonstrate the flexibility of our method on a diverse set of existing counting benchmarks: specifically cells, cars, and human crowds. The model achieves competitive performance on cell and crowd counting datasets, and surpasses the state-of-the-art on the car dataset using only three training images. When training on the entire dataset, the proposed method outperforms all previous methods by a large margin.
This paper introduces a new multi-modal dataset for visual and audio-visual speech recognition. It includes face tracks from over 400 hours of TED and TEDx videos, along with the corresponding subtitles and word alignment boundaries. The new dataset is substantially larger in scale compared to other public datasets that are available for general research.
The objective of this paper is to learn a compact representation of image sets for template-based face recognition. We make the following contributions: first, we propose a network architecture which aggregates and embeds the face descriptors produced by deep convolutional neural networks into a compact fixed-length representation. This compact representation requires minimal memory storage and enables efficient similarity computation. Second, we propose a novel GhostVLAD layer that includes {\em ghost clusters}, that do not contribute to the aggregation. We show that a quality weighting on the input faces emerges automatically such that informative images contribute more than those with low quality, and that the ghost clusters enhance the network's ability to deal with poor quality images. Third, we explore how input feature dimension, number of clusters and different training techniques affect the recognition performance. Given this analysis, we train a network that far exceeds the state-of-the-art on the IJB-B face recognition dataset. This is currently one of the most challenging public benchmarks, and we surpass the state-of-the-art on both the identification and verification protocols.