We consider the problem of localizing a novel image in a large 3D model, given that the gravitational vector is known. In principle, this is just an instance of camera pose estimation, but the scale of the problem introduces some interesting challenges. Most importantly, it makes the correspondence problem very difficult so there will often be a significant number of outliers to handle. To tackle this problem, we use recent theoretical as well as technical advances. Many modern cameras and phones have gravitational sensors that allow us to reduce the search space. Further, there are new techniques to efficiently and reliably deal with extreme rates of outliers. We extend these methods to camera pose estimation by using accurate approximations and fast polynomial solvers. Experimental results are given demonstrating that it is possible to reliably estimate the camera pose despite cases with more than 99 percent outlier correspondences in city-scale models with several millions of 3D points.

Abstract. We present a non-incremental approach to structure from motion. Our solution is based on robustly computing global rotations from relative geometries and feeding these into the known-rotation framework to create an initial solution for bundle adjustment. To increase robustness we present a new method for constructing reliable point tracks from pairwise matches. We show that our method can be seen as maximizing the reliability of a point track if the quality of the weakest link in the track is used to evaluate reliability. To estimate the final geometry we alternate between bundle adjustment and a robust version of the known-rotation formulation. The ability to compute both structure and camera translations independent of initialization makes our algorithm insensitive to degenerate epipolar geometries. We demonstrate the performance of our system on a number of image collections. 1

Prior work on multi-view structure from motion is dominated by sequential approaches starting from a single twoview reconstruction, then adding new images one by one. In contrast, we propose a non-sequential methodology based on rotational consistency and robust estimation using convex optimization. The resulting system is more robust with respect to (i) unreliable two-view estimations caused by short baselines, (ii) repetitive scenes with locally consistent structures that are not consistent with the global geometry and (iii) loop closing as errors are not propagated in a sequential manner. Both theoretical justifications and experimental comparisons are given to support these claims.

Correspondence problems are of great importance in computer vision. They appear as subtasks in many applications such as object recognition, merging partial 3D reconstructions and image alignment. Automatically matching features from appearance only is difficult and errors are frequent. Thus, it is necessary to use geometric consistency to remove incorrect correspondences. Typically heuristic methods like RANSAC or EM-like algorithms are used, but they risk getting trapped in local optima and are in no way guaranteed to find the best solution.This paper illustrates how pairwise constraints in combination with graph methods can be used to efficiently find optimal correspondences. These ideas are implemented on two basic geometric problems, 3D-3D registration and 2D-3D registration. The developed scheme can handle large rates of outliers and cope with multiple hypotheses. Despite the combinatorial explosion, the resulting algorithm which has been extensively evaluated on real data, yields competitive running times compared to state of the art.

We consider the problem of localizing a novel image in a large 3D model. In principle, this is just an instance of camera pose estimation, but the scale introduces some challenging problems. For one, it makes the correspondence problem very difficult and it is likely that there will be a significant rate of outliers to handle. In this paper we use recent theoretical as well as technical advances to tackle these problems. Many modern cameras and phones have gravitational sensors that allow us to reduce the search space. Further, there are new techniques to efficiently and reliably deal with extreme rates of outliers. We extend these methods to camera pose estimation by using accurate approximations and fast polynomial solvers. Experimental results are given demonstrating that it is possible to reliably estimate the camera pose despite more than 99% of outlier correspondences. 1

Background: Artificial intelligence (AI) is about to transform medical imaging. The Research Consortium for Medical Image Analysis (RECOMIA), a not-for-profit organisation, has developed an online platform to facilitate collaboration between medical researchers and AI researchers. The aim is to minimise the time and effort researchers need to spend on technical aspects, such as transfer, display, and annotation of images, as well as legal aspects, such as de-identification. The purpose of this article is to present the RECOMIA platform and its AI-based tools for organ segmentation in computed tomography (CT), which can be used for extraction of standardised uptake values from the corresponding positron emission tomography (PET) image. Results: The RECOMIA platform includes modules for (1) local de-identification of medical images, (2) secure transfer of images to the cloud-based platform, (3) display functions available using a standard web browser, (4) tools for manual annotation of organs or pathology in the images, (5) deep learning-based tools for organ segmentation or other customised analyses, (6) tools for quantification of segmented volumes, and (7) an export function for the quantitative results. The AI-based tool for organ segmentation in CT currently handles 100 organs (77 bones and 23 soft tissue organs). The segmentation is based on two convolutional neural networks (CNNs): one network to handle organs with multiple similar instances, such as vertebrae and ribs, and one network for all other organs. The CNNs have been trained using CT studies from 339 patients. Experienced radiologists annotated organs in the CT studies. The performance of the segmentation tool, measured as mean Dice index on a manually annotated test set, with 10 representative organs, was 0.93 for all foreground voxels, and the mean Dice index over the organs were 0.86 (0.82 for the soft tissue organs and 0.90 for the bones). Conclusion: The paper presents a platform that provides deep learning-based tools that can perform basic organ segmentations in CT, which can then be used to automatically obtain the different measurement in the corresponding PET image. The RECOMIA platform is available on request at www.recomia.org for research purposes.

The aim of this study was to develop a deep learning-based method for segmentation of bones in CT scans and test its accuracy compared to manual delineation, as a first step in the creation of an automated PET/ CT-based method for quantifying skeletal tumour burden. Methods: Convolutional neural networks (CNNs) were trained to segment 49 bones using manual segmentations from 100 CT scans. After training, the CNN-based segmentation method was tested on 46 patients with prostate cancer, who had undergone 18 F-choline-PET/CT and 18 F-NaF PET/CT less than three weeks apart. Bone volumes were calculated from the segmentations. The network's performance was compared with manual segmentations of five bones made by an experienced physician. Accuracy of the spatial overlap between automated CNN-based and manual segmentations of these five bones was assessed using the Sørensen-Dice index (SDI). Reproducibility was evaluated applying the Bland-Altman method. Results: The median (SD) volumes of the five selected bones were by CNN and manual segmentation: Th7 41 (3.8) and 36 (5.1), L3 76 (13) and 75 (9.2), sacrum 284 (40) and 283 (26), 7th rib 33 (3.9) and 31 (4.8), sternum 80 (11) and 72 (9.2), respectively. Median SDIs were 0.86 (Th7), 0.85 (L3), 0.88 (sacrum), 0.84 (7th rib) and 0.83 (sternum). The intraobserver volume difference was less with CNN-based than manual approach: Th7 2% and 14%, L3 7% and 8%, sacrum 1% and 3%, 7th rib 1% and 6%, sternum 3% and 5%, respectively. The average volume difference measured as ratio volume difference/mean volume between the two CNN-based segmentations was 5-6% for the vertebral column and ribs and ≤3% for other bones. Conclusion:The new deep learning-based method for automated segmentation of bones in CT scans provided highly accurate bone volumes in a fast and automated way and, thus, appears to be a valuable first step in the development of a clinical useful processing procedure providing reliable skeletal segmentation as a key part of quantification of skeletal metastases.

We present a framework for computing optimal transformations, aligning one point set to another, in the presence of outliers. Example applications include shape matching and registration (using, for example, similarity, affine or projective transformations) as well as multiview reconstruction problems (triangulation, camera pose etc.).While standard methods like RANSAC essentially use heuristics to cope with outliers, we seek to find the largest possible subset of consistent correspondences and the globally optimal transformation aligning the point sets. Based on theory from computational geometry, we show that this is indeed possible to accomplish in polynomial-time. We develop several algorithms which make efficient use of convex programming. The scheme has been tested and evaluated on both synthetic and real data for several applications. 1