No abstract

We study the problem of object recognition for categories for which we have no training examples, a task also called zero--data or zero-shot learning. This situation has hardly been studied in computer vision research, even though it occurs frequently; the world contains tens of thousands of different object classes, and image collections have been formed and suitably annotated for only a few of them. To tackle the problem, we introduce attribute-based classification: Objects are identified based on a high-level description that is phrased in terms of semantic attributes, such as the object's color or shape. Because the identification of each such property transcends the specific learning task at hand, the attribute classifiers can be prelearned independently, for example, from existing image data sets unrelated to the current task. Afterward, new classes can be detected based on their attribute representation, without the need for a new training phase. In this paper, we also introduce a new data set, Animals with Attributes, of over 30,000 images of 50 animal classes, annotated with 85 semantic attributes. Extensive experiments on this and two more data sets show that attribute-based classification indeed is able to categorize images without access to any training images of the target classes.

The increased availability of labeled X-ray image archives (e.g. ChestX-ray14 dataset) has triggered a growing interest in deep learning techniques. To provide better insight into the different approaches, and their applications to chest X-ray classification, we investigate a powerful network architecture in detail: the ResNet-50. Building on prior work in this domain, we consider transfer learning with and without fine-tuning as well as the training of a dedicated X-ray network from scratch. To leverage the high spatial resolution of X-ray data, we also include an extended ResNet-50 architecture, and a network integrating non-image data (patient age, gender and acquisition type) in the classification process. In a concluding experiment, we also investigate multiple ResNet depths (i.e. ResNet-38 and ResNet-101). In a systematic evaluation, using 5-fold re-sampling and a multi-label loss function, we compare the performance of the different approaches for pathology classification by ROC statistics and analyze differences between the classifiers using rank correlation. Overall, we observe a considerable spread in the achieved performance and conclude that the X-ray-specific ResNet-38, integrating non-image data yields the best overall results. Furthermore, class activation maps are used to understand the classification process, and a detailed analysis of the impact of non-image features is provided.

The optimization of k-space sampling for nonlinear sparse MRI reconstruction is phrased as a Bayesian experimental design problem. Bayesian inference is approximated by a novel relaxation to standard signal processing primitives, resulting in an efficient optimization algorithm for Cartesian and spiral trajectories. On clinical resolution brain image data from a Siemens 3T scanner, automatically optimized trajectories lead to significantly improved images, compared to standard low-pass, equispaced, or variable density randomized designs. Insights into the nonlinear design optimization problem for MRI are given. Modern MRI applications demand high imaging speed, especially when dynamic processes are observed, be it in cardiac examinations, angiography, or Functional MRI studies at high temporal resolutions. This need for rapid scanning can be dealt with by alternative encoding strategies, making use of multiple receiver coils (1-3) in order to parallelize the imaging process to some extent.While parallel MRI exploits redundancies between several receiver channels, imaging speed can also be increased by taking advantage of redundancies in the signal itself, which allows for reconstruction of the image from only a part of k-space in the first place. For example, k-space measurements show approximately Hermitian symmetry, which is exploited in partial Fourier acquisition techniques (4). Far beyond these simple symmetries, images form a statistically tightly constrained signal class. Fast, efficient digital image and video compression techniques are routinely used today, and the principles underlying them hold much promise for undersampled high-resolution MRI reconstruction (5-8), if this process is understood in terms of nonlinear statistical estimation. These proposals are now known as compressed sensing or sparse reconstruction since they exploit the statistical sparsity of images, a robust low-level characteristic, which leads to nonlinear yet conservative and well-characterized interpolation behavior (5). Compressed sensing is increasingly used for MRI problems, such as dynamic (9) imaging (10), as well as for spiral (11) and radial undersampling (12,13). Typically, scan time reductions by a factor of 2 or more can be achieved without losses in spatial resolution or sensitivity. Sparse statistics of images or image series originate from the structure of their pixel representations, an important instance of which is spatial or temporal redundancy, which has been used to speed up MRI acquisition (14-17).Two problems arise in practical applications of compressed sensing: how to reconstruct an image from a fixed undersampling design and how to choose the design in the first place. While a large amount of work was done for the former, we are not aware of much practically relevant progress for the latter. It is the undersampling design optimization problem for sparse MR image reconstruction that we focus on in this paper. Although there is substantial prior work on k-space optimization (18-20), this has been done for...