Learning and Understanding in 3D Point Clouds (Video1)Video Length:
We present a revised pipe-line of the existing 3D object detection and pose estimation framework based on point pair feature matching. In particular, we propose to couple the object detection with a coarse-to-fine segmentation, where each segment is subject to disjoint pose estimation. During matching, we apply a weighted Hough voting and an interpolated recovery of pose parameters. Finally, all the generated hypotheses are tested via an occlusion-aware ranking and sorted. We argue that such a combined pipeline simultaneously boosts the detection rate and reduces the complexity, while improving the accuracy of the resulting pose.
24 May 2019, MI 03.13.010 Invited Talk by Danielle Pace This work addresses the need for whole heart segmentation to individually label the cardiac chambers and great vessels for patients with congenital heart disease. Our main challenges are extreme anatomical variability and very limited training data. We demonstrate an iterative segmentation method, implemented as a recurrent neural network (RNN), in which a user provides a single landmark per structure, and a segmentation is evolved over multiple steps until reaching a stopping point that can be user-defined or automatically determined. The model grows segmentations in a predictable way that is defined during training. We show that a loss function that evaluates the entire sequence of output segmentations can be optimized using training images alongside input-output pairs of partial segmentations. Our experiments demonstrate that, compared to conventional models that segment an image in one step, the iterative segmentation offers better generalization to patients with the most severe heart malformations.
17 January 2019, MI HS 2 Invited Talk by Prof. Oliver Bimber Digital images play an essential role in our life. Advanced imaging systems, image processing methods, and visualization techniques are today fundamental to many professions. Medical imaging is certainly a good example. However, when mapping complex (possibly multidimensional) data to 2D, information is lost. What if the notion of digital images would change once and forever? What if instead of capturing, storing, processing and displaying only a single color per pixel, each pixel would consist of individual colors for each emitting direction? Digital images would no longer be two-dimensional matrices but four-dimensional ones (storing spatial information in two dimensions, and directional information in the other two dimensions). This is called a light field. I will introduce the fundamentals of light fields, explain how light fields are captured, processed, and displayed, and present several applications of light-field technology in various application domains, such as microscopy, sensors, and aerial scanning.
6 July 2018, MI 03.13.010 Invited Talk by Prof. William (Sandy) Wells Segmentation is a fundamental task for extracting semantically meaningful regions from an image. The goal of segmentation algorithms is to accurately assign object labels to each image location. However, image-noise, shortcomings of algorithms, and image ambiguities cause uncertainty in label assignment. Estimating the uncertainty in label assignment is important in multiple application domains, such as segmenting tumors from medical images for radiation treatment planning. One way to estimate these uncertainties is through the computation of posteriors of Bayesian models, which is computationally prohibitive for many practical applications. On the other hand, most computationally efficient methods fail to estimate label uncertainty. [...]
28 February 2019, CAMP/CM@TUM offers up to 5 International Undergraduate Excellence Awards in 2019: to the best international undergraduate applicants The Chair for Computer Aided Medical Procedures (CAMP) and the Chair for Connected Mobility would like to offer up to 5 research awards to strongly motivated international undergraduate students interested in improving their research skills in machine learning, medical imaging, computer vision, augmented reality, Internet of Things, and/or Edge Computing. The students are required to pursue their undergraduate studies outside Germany. The first two awards amount to 4500 Euros and the second three awards to 4000 Euros. These awards will provide financial support for the awardees during their research stay at CAMP for a period of up to three months between June 2019 and October 2019.
19 December 2018, 00.12.019 Geometric Methods for 3D Reconstruction from Large Point Clouds 3D reconstruction involves the task of capturing the shape and appearance of objects or scenes in the formof 3D computer aided design (CAD) models, often throughmultiple measurements: individual 2D images or multiview 3D data. In this thesis, we explore both sparse and dense 3D reconstruction methods in scenarios where 3D cues are present and rough, prior CAD models are at hand before the operation. Thanks to the proliferation of 3D sensors and increased accessibility of 3D data, we can now remain true to the 3D nature of our physical world in such digitization processes and utilize direct 3D input, point clouds, which can also alleviate problems of capture modality and illumination conditions. Both sparse and dense reconstruction problems arise frequently either in industrial machine vision where the production processes of parts and goods are to be inspected, or in restoration applications where crude digital models are desired to be improved. We start by explaining our contributions to sparse yet accurate reconstruction from non-overlapping multiview images. The experience developed here have also been used to acquire accurate ground truth aiding the assessment of the next stage, dense reconstruction.We tackle the latter by proposing a novel, multiview point cloud based 3D reconstruction pipeline in which it is possible to incorporate CAD proxies. This is accomplished by first aligning all the scans not by an O(N2) inter-scan matching but by a linear scan-to-model registration. Such alignment ismade possible by novel object detection and pose estimation algorithms.Next, respecting the deviations of the real data from the CAD model, we performa CAD-free multi-scan refinement further increasing the accuracy both qualitatively and quantitatively.We also propose novel methods to initialize such large scale optimization problems and to infer information about the reliability of solutions, known as the uncertainty. As seen, the necessity to incorporate the CAD models as proxies to reconstruction comes along withmany challenges to be addressed. In this thesis, the prominent subtasks include preparing CAD models towards the reconstruction task, object detection, estimation of full six degree of freedom (DoF?) rigid pose and pose graph optimization, in which the roughly aligned scans are brought to the final alignment.We address all of those problems with rigor.Moreover, thanks to our photogrammetric ground truth acquisition strategies, we present thorough evaluation of all tasks, in real datasets besides synthetic ones.
6 December 2018, 00.12.019 PhD Defense by Christian Rupprecht Nearly all real-world image understanding problems in computer vision are inherently ambiguous. Often, predictive systems do not model this ambiguity and do not consider the possibility that there can be more than just a single outcome for a given problem. This leads to sub-par performance on ambiguous tasks as the model has to account for all possibilities with one answer. We define three typical sources of confusion that render tasks not optimally solvable with a single unique prediction. In this dissertation we describe two principled and general approaches of dealing with ambiguity. First, we elaborate on a method that allows the algorithm to predict multiple instead of one single answer. This is a pragmatic way of dealing with ambiguity: instead of deciding for an exclusive outcome for a given problem, we enable the system to list several possibilities. The second part describes an alternative way to deal with uncertain predictions. Often human perception can provide additional information about a task or application that an intelligent system might have not recognized. Building on the paradigm of human-machine interaction, we show how enabling interaction between the system and a user can improve predictions on the example of semantic segmentation. We describe a novel guiding mechanism that can be seamlessly integrated into the system and shows great potential beyond the demonstrated tasks for several further applications.
19 November 2018, 01.09.014 PhD Defense by Nicola Rieke Visual tracking of surgical instruments is a key component of various computer-assisted interventions, yet a very challenging problem in the field of Computer Vision. This dissertation presents novel approaches which leverage machine learning techniques for precise real-time tracking and 2D pose estimation of instruments. The achieved results demonstrate that the proposed methods based on Random Forests and Deep Learning provide remarkable advantages with respect to the state of the art in terms of accuracy, robustness and generalization.
8 October 2018, 01.06.020 Signed Distance Fields for Rigid and Deformable 3D Reconstruction Capturing three-dimensional environments is a key task in the growing fields of virtual and augmented reality. This thesis addresses the task of 3D reconstruction of both static and dynamic objects and scenes scanned with a single hand-held RGB-D camera, without any markers or prior knowledge. Reconstructing rigid environments requires estimating the six degrees-of-freedom camera pose at every time instance, and subsequently fusing the acquired data into a geometrically consistent computer model. The task of reconstructing deformable objects is more challenging, as additionally the non-rigid motion that occurred in every frame has to be determined and factored out. We propose to tackle both the rigid and deformable reconstruction problems via implicit-to-implicit alignment of SDF pairs without correspondence search. In the static case, we obtain more accurate pose estimates with a framework that permits straightforward incorporation of various additional constraints, such as surface colour and orientation. We start with the reconstruction of small- to medium-scale household objects and demonstrate how to extend the approach to larger spaces such as rooms. To this end, we develop a limited-extent volume strategy that restricts registration to the most geometrically distinctive regions of a scene, leading to significantly improved rotational motion estimation. Finally, we adapt our approach to dynamic scenes by modifying our implicit-to-implicit approach so that new data is incremented appropriately. For this purpose we evolve an initial SDF to a target SDF by imposing rigidity constraints that require the underlying deformation field to be approximately Killing, i.e. volume-preserving and generating locally isometric motions. Alternatively, we employ gradient flow in the smooth Sobolev space, which favours global deformations over finer-scale details. These strategies also circumvent explicit correspondence search and thus avoid the repeated conversion between SDF and mesh representations that other techniques entail. Nevertheless, we ensure that correspondence information can be recovered by proposing two strategies based on Laplacian eigenfunctions, which are known to encode natural deformation patterns. Thanks to the used SDF representation, our non-rigid reconstruction approach is able to handle topological changes and fast motion, which are major obstacles for existing approaches.
28 February 2019, London, UK 1 paper accepted at MIDL 2019 We are happy to announce that one paper from our Chair will be presented at the International Conference on Medical Imaging with Deep Learning MIDL 2019, which will take place in London, July 8-9, 2019.
26 February 2019, Hong Kong 2 papers accepted at IPMI 2019 We are happy to announce that two papers from our Chair will be presented at the International Conference on Information Processing in Medical Imaging, IPMI 2019, which will take place in Hong Kong, June 2-7, 2019.
25 February 2019, Long Beach, CA, USA 5 papers accepted at CVPR 2019 We are happy to announce that five papers from our Chair will be presented at the International Conference on Computer Vision and Pattern Recognition, CVPR 2019, which will take place in Long Beach, CA, USA on June 16-20, 2019.
1 February 2019, Venice, Italy 2 papers accepted at ISBI 2019 We are happy to announce that two papers from our Chair will be presented at the International Symposium on Biomedical Imaging, ISBI 2019, which will take place in Venice, Italy on April 8-11, 2019.
3 December 2018, Montréal Canada 1 paper accepted at NIPS 2018 We are happy to announce that 1 paper from our chair has been published at NIPS 2018.