Models, code, and papers for "Alexey Dosovitskiy":
We address the problem of learning accurate 3D shape and camera pose from a collection of unlabeled category-specific images. We train a convolutional network to predict both the shape and the pose from a single image by minimizing the reprojection error: given several views of an object, the projections of the predicted shapes to the predicted camera poses should match the provided views. To deal with pose ambiguity, we introduce an ensemble of pose predictors which we then distill to a single "student" model. To allow for efficient learning of high-fidelity shapes, we represent the shapes by point clouds and devise a formulation allowing for differentiable projection of these. Our experiments show that the distilled ensemble of pose predictors learns to estimate the pose accurately, while the point cloud representation allows to predict detailed shape models. The supplementary video can be found at https://www.youtube.com/watch?v=LuIGovKeo60
We present an approach to sensorimotor control in immersive environments. Our approach utilizes a high-dimensional sensory stream and a lower-dimensional measurement stream. The cotemporal structure of these streams provides a rich supervisory signal, which enables training a sensorimotor control model by interacting with the environment. The model is trained using supervised learning techniques, but without extraneous supervision. It learns to act based on raw sensory input from a complex three-dimensional environment. The presented formulation enables learning without a fixed goal at training time, and pursuing dynamically changing goals at test time. We conduct extensive experiments in three-dimensional simulations based on the classical first-person game Doom. The results demonstrate that the presented approach outperforms sophisticated prior formulations, particularly on challenging tasks. The results also show that trained models successfully generalize across environments and goals. A model trained using the presented approach won the Full Deathmatch track of the Visual Doom AI Competition, which was held in previously unseen environments.
Feature representations, both hand-designed and learned ones, are often hard to analyze and interpret, even when they are extracted from visual data. We propose a new approach to study image representations by inverting them with an up-convolutional neural network. We apply the method to shallow representations (HOG, SIFT, LBP), as well as to deep networks. For shallow representations our approach provides significantly better reconstructions than existing methods, revealing that there is surprisingly rich information contained in these features. Inverting a deep network trained on ImageNet provides several insights into the properties of the feature representation learned by the network. Most strikingly, the colors and the rough contours of an image can be reconstructed from activations in higher network layers and even from the predicted class probabilities.
Image-generating machine learning models are typically trained with loss functions based on distance in the image space. This often leads to over-smoothed results. We propose a class of loss functions, which we call deep perceptual similarity metrics (DeePSiM), that mitigate this problem. Instead of computing distances in the image space, we compute distances between image features extracted by deep neural networks. This metric better reflects perceptually similarity of images and thus leads to better results. We show three applications: autoencoder training, a modification of a variational autoencoder, and inversion of deep convolutional networks. In all cases, the generated images look sharp and resemble natural images.
Navigation research is attracting renewed interest with the advent of learning-based methods. However, this new line of work is largely disconnected from well-established classic navigation approaches. In this paper, we take a step towards coordinating these two directions of research. We set up classic and learning-based navigation systems in common simulated environments and thoroughly evaluate them in indoor spaces of varying complexity, with access to different sensory modalities. Additionally, we measure human performance in the same environments. We find that a classic pipeline, when properly tuned, can perform very well in complex cluttered environments. On the other hand, learned systems can operate more robustly with a limited sensor suite. Overall, both approaches are still far from human-level performance.
Manually re-drawing an image in a certain artistic style takes a professional artist a long time. Doing this for a video sequence single-handedly is beyond imagination. We present two computational approaches that transfer the style from one image (for example, a painting) to a whole video sequence. In our first approach, we adapt to videos the original image style transfer technique by Gatys et al. based on energy minimization. We introduce new ways of initialization and new loss functions to generate consistent and stable stylized video sequences even in cases with large motion and strong occlusion. Our second approach formulates video stylization as a learning problem. We propose a deep network architecture and training procedures that allow us to stylize arbitrary-length videos in a consistent and stable way, and nearly in real time. We show that the proposed methods clearly outperform simpler baselines both qualitatively and quantitatively. Finally, we propose a way to adapt these approaches also to 360 degree images and videos as they emerge with recent virtual reality hardware.
We introduce a new memory architecture for navigation in previously unseen environments, inspired by landmark-based navigation in animals. The proposed semi-parametric topological memory (SPTM) consists of a (non-parametric) graph with nodes corresponding to locations in the environment and a (parametric) deep network capable of retrieving nodes from the graph based on observations. The graph stores no metric information, only connectivity of locations corresponding to the nodes. We use SPTM as a planning module in a navigation system. Given only 5 minutes of footage of a previously unseen maze, an SPTM-based navigation agent can build a topological map of the environment and use it to confidently navigate towards goals. The average success rate of the SPTM agent in goal-directed navigation across test environments is higher than the best-performing baseline by a factor of three. A video of the agent is available at https://youtu.be/vRF7f4lhswo
We present a deep convolutional decoder architecture that can generate volumetric 3D outputs in a compute- and memory-efficient manner by using an octree representation. The network learns to predict both the structure of the octree, and the occupancy values of individual cells. This makes it a particularly valuable technique for generating 3D shapes. In contrast to standard decoders acting on regular voxel grids, the architecture does not have cubic complexity. This allows representing much higher resolution outputs with a limited memory budget. We demonstrate this in several application domains, including 3D convolutional autoencoders, generation of objects and whole scenes from high-level representations, and shape from a single image.
In the past, manually re-drawing an image in a certain artistic style required a professional artist and a long time. Doing this for a video sequence single-handed was beyond imagination. Nowadays computers provide new possibilities. We present an approach that transfers the style from one image (for example, a painting) to a whole video sequence. We make use of recent advances in style transfer in still images and propose new initializations and loss functions applicable to videos. This allows us to generate consistent and stable stylized video sequences, even in cases with large motion and strong occlusion. We show that the proposed method clearly outperforms simpler baselines both qualitatively and quantitatively.
We present a convolutional network capable of inferring a 3D representation of a previously unseen object given a single image of this object. Concretely, the network can predict an RGB image and a depth map of the object as seen from an arbitrary view. Several of these depth maps fused together give a full point cloud of the object. The point cloud can in turn be transformed into a surface mesh. The network is trained on renderings of synthetic 3D models of cars and chairs. It successfully deals with objects on cluttered background and generates reasonable predictions for real images of cars.
Latest results indicate that features learned via convolutional neural networks outperform previous descriptors on classification tasks by a large margin. It has been shown that these networks still work well when they are applied to datasets or recognition tasks different from those they were trained on. However, descriptors like SIFT are not only used in recognition but also for many correspondence problems that rely on descriptor matching. In this paper we compare features from various layers of convolutional neural nets to standard SIFT descriptors. We consider a network that was trained on ImageNet and another one that was trained without supervision. Surprisingly, convolutional neural networks clearly outperform SIFT on descriptor matching. This paper has been merged with arXiv:1406.6909
When deep learning is applied to visual object recognition, data augmentation is often used to generate additional training data without extra labeling cost. It helps to reduce overfitting and increase the performance of the algorithm. In this paper we investigate if it is possible to use data augmentation as the main component of an unsupervised feature learning architecture. To that end we sample a set of random image patches and declare each of them to be a separate single-image surrogate class. We then extend these trivial one-element classes by applying a variety of transformations to the initial 'seed' patches. Finally we train a convolutional neural network to discriminate between these surrogate classes. The feature representation learned by the network can then be used in various vision tasks. We find that this simple feature learning algorithm is surprisingly successful, achieving competitive classification results on several popular vision datasets (STL-10, CIFAR-10, Caltech-101).
In dynamic environments, learned controllers are supposed to take motion into account when selecting the action to be taken. However, in existing reinforcement learning works motion is rarely treated explicitly; it is rather assumed that the controller learns the necessary motion representation from temporal stacks of frames implicitly. In this paper, we show that for continuous control tasks learning an explicit representation of motion improves the quality of the learned controller in dynamic scenarios. We demonstrate this on common benchmark tasks (Walker, Swimmer, Hopper), on target reaching and ball catching tasks with simulated robotic arms, and on a dynamic single ball juggling task. Moreover, we find that when equipped with an appropriate network architecture, the agent can, on some tasks, learn motion features also with pure reinforcement learning, without additional supervision. Further we find that using an image difference between the current and the previous frame as an additional input leads to better results than a temporal stack of frames.
End-to-end approaches to autonomous driving have high sample complexity and are difficult to scale to realistic urban driving. Simulation can help end-to-end driving systems by providing a cheap, safe, and diverse training environment. Yet training driving policies in simulation brings up the problem of transferring such policies to the real world. We present an approach to transferring driving policies from simulation to reality via modularity and abstraction. Our approach is inspired by classic driving systems and aims to combine the benefits of modular architectures and end-to-end deep learning approaches. The key idea is to encapsulate the driving policy such that it is not directly exposed to raw perceptual input or low-level vehicle dynamics. We evaluate the presented approach in simulated urban environments and in the real world. In particular, we transfer a driving policy trained in simulation to a 1/5-scale robotic truck that is deployed in a variety of conditions, with no finetuning, on two continents. The supplementary video can be viewed at https://youtu.be/BrMDJqI6H5U
Our understanding of reinforcement learning (RL) has been shaped by theoretical and empirical results that were obtained decades ago using tabular representations and linear function approximators. These results suggest that RL methods that use temporal differencing (TD) are superior to direct Monte Carlo estimation (MC). How do these results hold up in deep RL, which deals with perceptually complex environments and deep nonlinear models? In this paper, we re-examine the role of TD in modern deep RL, using specially designed environments that control for specific factors that affect performance, such as reward sparsity, reward delay, and the perceptual complexity of the task. When comparing TD with infinite-horizon MC, we are able to reproduce classic results in modern settings. Yet we also find that finite-horizon MC is not inferior to TD, even when rewards are sparse or delayed. This makes MC a viable alternative to TD in deep RL.
Autonomous driving models should ideally be evaluated by deploying them on a fleet of physical vehicles in the real world. Unfortunately, this approach is not practical for the vast majority of researchers. An attractive alternative is to evaluate models offline, on a pre-collected validation dataset with ground truth annotation. In this paper, we investigate the relation between various online and offline metrics for evaluation of autonomous driving models. We find that offline prediction error is not necessarily correlated with driving quality, and two models with identical prediction error can differ dramatically in their driving performance. We show that the correlation of offline evaluation with driving quality can be significantly improved by selecting an appropriate validation dataset and suitable offline metrics. The supplementary video can be viewed at https://www.youtube.com/watch?v=P8K8Z-iF0cY
We train generative 'up-convolutional' neural networks which are able to generate images of objects given object style, viewpoint, and color. We train the networks on rendered 3D models of chairs, tables, and cars. Our experiments show that the networks do not merely learn all images by heart, but rather find a meaningful representation of 3D models allowing them to assess the similarity of different models, interpolate between given views to generate the missing ones, extrapolate views, and invent new objects not present in the training set by recombining training instances, or even two different object classes. Moreover, we show that such generative networks can be used to find correspondences between different objects from the dataset, outperforming existing approaches on this task.
Most modern convolutional neural networks (CNNs) used for object recognition are built using the same principles: Alternating convolution and max-pooling layers followed by a small number of fully connected layers. We re-evaluate the state of the art for object recognition from small images with convolutional networks, questioning the necessity of different components in the pipeline. We find that max-pooling can simply be replaced by a convolutional layer with increased stride without loss in accuracy on several image recognition benchmarks. Following this finding -- and building on other recent work for finding simple network structures -- we propose a new architecture that consists solely of convolutional layers and yields competitive or state of the art performance on several object recognition datasets (CIFAR-10, CIFAR-100, ImageNet). To analyze the network we introduce a new variant of the "deconvolution approach" for visualizing features learned by CNNs, which can be applied to a broader range of network structures than existing approaches.