Attend and Segment: Attention Guided Active Semantic Segmentation

In a dynamic environment, an agent with a limited field of view/resource cannot fully observe the scene before attempting to parse it. The deployment of common semantic segmentation architectures is not feasible in such settings. In this paper we propose a method to gradually segment a scene given a sequence of partial observations. The main idea is to refine an agent's understanding of the environment by attending the areas it is most uncertain about. Our method includes a self-supervised attention mechanism and a specialized architecture to maintain and exploit spatial memory maps for filling-in the unseen areas in the environment. The agent can select and attend an area while relying on the cues coming from the visited areas to hallucinate the other parts. We reach a mean pixel-wise accuracy of 78.1%, 80.9% and 76.5% on CityScapes, CamVid, and Kitti datasets by processing only 18% of the image pixels (10 retina-like glimpses). We perform an ablation study on the number of glimpses, input image size and effectiveness of retina-like glimpses. We compare our method to several baselines and show that the optimal results are achieved by having access to a very low resolution view of the scene at the first timestep.

MIX'EM: Unsupervised Image Classification using a Mixture of Embeddings

We present MIX'EM, a novel solution for unsupervised image classification. Our model generates representations that by themselves are sufficient to drive a general-purpose clustering method to deliver high-quality classification without supervision. MIX'EM integrates an internal mixture of embeddings module into the contrastive visual representation learning framework to disentangle the representation space at the category level. It generates a set of embeddings from a visual representation and mixes them to construct the contrastive loss input. Parallel to the contrastive loss, we introduce three techniques to train MIX'EM and avoid a degenerate solution; (i) we maximize entropy across mixture components to diversify them, and (ii) minimize component entropy conditioned on instances to enforce a clustered embedding space. Applying (i) and (ii) lead to the emergence of semantic categories through the mixture coefficients, making it possible to (iii) apply an associative embedding loss to enforce semantic separability directly. Subsequently, we run K-means on the representations to acquire semantic classification, which outperforms the state-of-the-art by a large margin. We conduct extensive experiments and analyses on STL10, CIFAR10, and CIFAR100-20 datasets, achieving 78\%, 82\%, and 44\% accuracy, respectively. Essential to robust high accuracy is using MIX'EM to initialize K-means. Finally, we report impressively high accuracy baselines (70\% on STL10) achieved solely by applying K-means to the "normalized" representations learned using the contrastive loss.

Automatic Recall Machines: Internal Replay, Continual Learning and the Brain

Replay in neural networks involves training on sequential data with memorized samples, which counteracts forgetting of previous behavior caused by non-stationarity. We present a method where these auxiliary samples are generated on the fly, given only the model that is being trained for the assessed objective, without extraneous buffers or generator networks. Instead the implicit memory of learned samples within the assessed model itself is exploited. Furthermore, whereas existing work focuses on reinforcing the full seen data distribution, we show that optimizing for not forgetting calls for the generation of samples that are specialized to each real training batch, which is more efficient and scalable. We consider high-level parallels with the brain, notably the use of a single model for inference and recall, the dependency of recalled samples on the current environment batch, top-down modulation of activations and learning, abstract recall, and the dependency between the degree to which a task is learned and the degree to which it is recalled. These characteristics emerge naturally from the method without being controlled for.

Unsupervised Model Personalization while Preserving Privacy and Scalability: An Open Problem

This work investigates the task of unsupervised model personalization, adapted to continually evolving, unlabeled local user images. We consider the practical scenario where a high capacity server interacts with a myriad of resource-limited edge devices, imposing strong requirements on scalability and local data privacy. We aim to address this challenge within the continual learning paradigm and provide a novel Dual User-Adaptation framework (DUA) to explore the problem. This framework flexibly disentangles user-adaptation into model personalization on the server and local data regularization on the user device, with desirable properties regarding scalability and privacy constraints. First, on the server, we introduce incremental learning of task-specific expert models, subsequently aggregated using a concealed unsupervised user prior. Aggregation avoids retraining, whereas the user prior conceals sensitive raw user data, and grants unsupervised adaptation. Second, local user-adaptation incorporates a domain adaptation point of view, adapting regularizing batch normalization parameters to the user data. We explore various empirical user configurations with different priors in categories and a tenfold of transforms for MIT Indoor Scene recognition, and classify numbers in a combined MNIST and SVHN setup. Extensive experiments yield promising results for data-driven local adaptation and elicit user priors for server adaptation to depend on the model rather than user data. Hence, although user-adaptation remains a challenging open problem, the DUA framework formalizes a principled foundation for personalizing both on server and user device, while maintaining privacy and scalability.

Deep-Geometric 6 DoF Localization from a Single Image in Topo-metric Maps

We describe a Deep-Geometric Localizer that is able to estimate the full 6 Degree of Freedom (DoF) global pose of the camera from a single image in a previously mapped environment. Our map is a topo-metric one, with discrete topological nodes whose 6 DoF poses are known. Each topo-node in our map also comprises of a set of points, whose 2D features and 3D locations are stored as part of the mapping process. For the mapping phase, we utilise a stereo camera and a regular stereo visual SLAM pipeline. During the localization phase, we take a single camera image, localize it to a topological node using Deep Learning, and use a geometric algorithm (PnP) on the matched 2D features (and their 3D positions in the topo map) to determine the full 6 DoF globally consistent pose of the camera. Our method divorces the mapping and the localization algorithms and sensors (stereo and mono), and allows accurate 6 DoF pose estimation in a previously mapped environment using a single camera. With potential VR/AR and localization applications in single camera devices such as mobile phones and drones, our hybrid algorithm compares favourably with the fully Deep-Learning based Pose-Net that regresses pose from a single image in simulated as well as real environments.

Information Compensation for Deep Conditional Generative Networks

In recent years, unsupervised/weakly-supervised conditional generative adversarial networks (GANs) have achieved many successes on the task of modeling and generating data. However, one of their weaknesses lies in their poor ability to separate, or disentangle, the different factors that characterize the representation encoded in their latent space. To address this issue, we propose a novel structure for unsupervised conditional GANs powered by a novel Information Compensation Connection (IC-Connection). The proposed IC-Connection enables GANs to compensate for information loss incurred during deconvolution operations. In addition, to quantify the degree of disentanglement on both discrete and continuous latent variables, we design a novel evaluation procedure. Our empirical results suggest that our method achieves better disentanglement compared to the state-of-the-art GANs in a conditional generation setting.

Ternary Feature Masks: continual learning without any forgetting

In this paper, we propose an approach without any forgetting to continual learning for the task-aware regime, where at inference the task-label is known. By using ternary masks we can upgrade a model to new tasks, reusing knowledge from previous tasks while not forgetting anything about them. Using masks prevents both catastrophic forgetting and backward transfer. We argue -- and show experimentally -- that avoiding the former largely compensates for the lack of the latter, which is rarely observed in practice. In contrast to earlier works, our masks are applied to the features (activations) of each layer instead of the weights. This considerably reduces the number of mask parameters to be added for each new task; with more than three orders of magnitude for most networks. The encoding of the ternary masks into two bits per feature creates very little overhead to the network, avoiding scalability issues. Our masks do not permit any changes to features which are used by previous tasks. As this may be too restrictive to allow learning of new tasks, we add task-specific feature normalization. This way, already learned features can adapt to the current task without changing the behavior of these features for previous tasks. Extensive experiments on several finegrained datasets and ImageNet show that our method outperforms current state-of-the-art while reducing memory overhead in comparison to weight-based approaches.

Dynamic Convolutions: Exploiting Spatial Sparsity for Faster Inference

Modern convolutional neural networks apply the same operations on every pixel in an image. However, not all image regions are equally important. To address this inefficiency, we propose a method to dynamically apply convolutions conditioned on the input image. We introduce a residual block where a small gating branch learns which spatial positions should be evaluated. These discrete gating decisions are trained end-to-end using the Gumbel-Softmax trick, in combination with a sparsity criterion. Our experiments on Food-101, CIFAR and ImageNet show that our method has better focus on the region of interest and better accuracy than existing methods, at a lower computational complexity. Moreover, we provide an efficient CUDA implementation of our dynamic convolutions using a gather-scatter approach, achieving a significant improvement in inference speed on MobileNetV2 and ShuffleNetV2. On human pose estimation, a task that is inherently spatially sparse, the processing speed is increased by 45% with less than 0.1% loss in accuracy.

Mixture Dense Regression for Object Detection and Human Pose Estimation

Mixture models are well-established machine learning approaches that, in computer vision, have mostly been applied to inverse or ill-defined problems. However, they are general-purpose divide-and-conquer techniques, splitting the input space into relatively homogeneous subsets, in a data-driven manner. Therefore, not only ill-defined but also well-defined complex problems should benefit from them. To this end, we devise a multi-modal solution for spatial regression using mixture density networks for dense object detection and human pose estimation. For both tasks, we show that a mixture model converges faster, yields higher accuracy, and divides the input space into interpretable modes. For object detection, mixture components learn to focus on object scale with the distribution of components closely following the distribution of ground truth object scale. For human pose estimation, a mixture model divides the data based on viewpoint and uncertainty -- namely, front and back views, with back view imposing higher uncertainty. We conduct our experiments on the MS COCO dataset and do not face any mode collapse. However, to avoid numerical instabilities, we had to modify the activation function for the mixture variance terms slightly.

How to improve CNN-based 6-DoF camera pose estimation

Convolutional neural networks (CNNs) and transfer learning have recently been used for 6 degrees of freedom (6-DoF) camera pose estimation. While they do not reach the same accuracy as visual SLAM-based approaches and are restricted to a specific environment, they excel in robustness and can be applied even to a single image. In this paper, we study PoseNet [1] and investigate modifications based on datasets' characteristics to improve the accuracy of the pose estimates. In particular, we emphasize the importance of field-of-view over image resolution; we present a data augmentation scheme to reduce overfitting; we study the effect of Long-Short-Term-Memory (LSTM) cells. Lastly, we combine these modifications and improve PoseNet's performance for monocular CNN based camera pose regression.