A major bottleneck in search-based program synthesis, which learns programs from input/output examples, is the synthesis of large programs. As the size of the target program increases, so does the search depth, which leads to an exponentially growing number of candidate programs. Humans mitigate the combinatorial explosion that arises from deep program search: they build complex programs from smaller parts. We introduce a new strategy for program synthesis called Divide, Align & Conquer (DA&C) that exploits the compositionality of real-world domains to guide the synthesis towards useful subprograms. Divide decomposes each example using a segmentation procedure that is synthesized as part of the learning problem. Align matches the components in the decomposed input/output examples to steer the search toward combinations that lead to the synthesis of useful subprograms, and Conquer then solves a standalone synthesis problem on each pair of aligned input/output components. We show how replacing a deep program search with a linear number of much smaller synthesis tasks leads us to efficiently discover useful subprograms that are then combined into a solution program. Our agent outperforms current Inductive Logic Programming (ILP) methods on string transformation tasks even with minimal knowledge priors. Unlike existing methods, the predictive accuracy of our agent monotonically increases for additional examples. It approximates an average time complexity of O(m) in the size m of subprograms for highly structured and, hence, decomposable domains such as strings. Finally, we demonstrate the scalability of our technique on highdimensional abstract visual reasoning tasks from the Abstract Reasoning Corpus (ARC) for which ILP methods were previously infeasible. We are competitive with state-of-the-art agents outside of ILP, despite generating only 0.2% as many candidate programs from a knowledge prior of only 11 generic geometric primitives.

Despite remarkable advances in the development of language resources over the recent years, there is still a shortage of annotated, publicly available corpora covering (German) medical language. With the initial release of the German Guideline Program in Oncology NLP Corpus (GGPONC), we have demonstrated how such corpora can be built upon clinical guidelines, a widely available resource in many natural languages with a reasonable coverage of medical terminology. In this work, we describe a major new release for GGPONC. The corpus has been substantially extended in size and re-annotated with a new annotation scheme based on SNOMED CT top level hierarchies, reaching high inter-annotator agreement (γ=.94). Moreover, we annotated elliptical coordinated noun phrases and their resolutions, a common language phenomenon in (not only German) scientific documents. We also trained BERT-based named entity recognition models on this new data set, which achieve high performance on short, coarse-grained entity spans (F1=.89), while the rate of boundary errors increases for long entity spans. GGPONC is freely available through a data use agreement. The trained named entity recognition models, as well as the detailed annotation guide, are also made publicly available.

Reliable lane detection is a fundamental necessity for driver assistance, driver safety functions and fully automated vehicles. Based on other detection and classification tasks, deep learning based methods are likely to yield the most accurate outputs for detecting lane markers, but require vast amounts of labeled data. We propose to train a deep neural network for detecting lane markers in camera images without manually labeling any images. To achieve this, we project high definition maps for automated driving into our image and correct for misalignments due to inaccuracies in localization and coordinate frame transformations. The corrections are performed by calculating the offset between features within our map and detected ones in the images. By using detections in the actual image for refining the projections, our labels become close to pixel perfect. After a fast, visual quality check, our projected lane markers can be used for training a fully convolutional network to segment lane markers in images. A single worker can easily generate 20,000 of those labels within a single day. Our fully convolutional network is trained only on automatically generated labels. All of our detections are based solely on gray-scale mono camera inputs without any additional information. The resulting network regularly detects clean lane markers at distances of around 150 meters on a 1 Megapixel camera.

In this paper, we propose a novel multi-view method for surface reconstruction from matched line segments with applications to robotic mapping and image-based rendering. Starting from 3D line segments, plane hypotheses are formed for all non-collinear and sufficiently coplanar segment pairs. The surface that is spanned by two segments is used to immediately prune hypotheses that do not pass a sight line occlusion check to keep the initial plane number tractable. After further merging, exhaustive intersections are computed in an efficient way to yield a maximally informative surface representation. Finally, robustified visibility constraints are used to recover a dense surface mesh that is a pessimistic representation of the free space, which is desirable for path planning applications. The presented system is a complete and automatic solution suitable for mapping an environment in realtime scenarios like robotic exploration. We demonstrate the performance of our algorithm on several indoor scenes with varying complexity.

This work is concerned with the matching of straight lines between two stereo image pairs by reprojection. While we will focus on visual odometry in the realm of simultaneous mapping and localization, the techniques are also relevant to monocular and stereo 3D object detection and tracking. Our first contribution is an adaptation of the Iterative Closest Point (ICP) algorithm to the domain of lines. We argue that a naive “Iterative Closest Line” derivation cannot deliver similar performance. In contrast, our novel Iterative Closest Multiple Lines (ICML) algorithm allows efficient line matching while even reducing the amount of local minima during iterative optimization with its consideration of several weighted matches. The second contribution is a fast and robust hypothesize-and-test algorithm which can act as a fallback for challenging frame pairs where pure gradient-based optimization fails. In several differently textured scenes, we demonstrate robust performance, even in very sparse cases where proven feature point based methods fail. In comparison to edge-point ICP, we see speed improvements of more than an order of a magnitude and reduced susceptibility for local minima.