One longstanding challenge in the field of robotics has been the robust and reliable grasping of objects of unknown shape. Part of that challenge lies in reconstructing the object’s shape using only limited observations. Most approaches use either visual or tactile information to reconstruct the shape, having to face issues resulting from the limitations of the chosen modality. This thesis tries to combine the strengths of visual and tactile observations by taking the result from an existing visual approach and refining that result through sparse tactile glances. The existing approach produces potential shape hypotheses in voxel space which get combined into one final shape. This thesis takes that final shape and determines voxels of interest using either entropy or variance. These voxels will be targeted by the exploration, providing information about these voxels. This information will be used to assign weights to the original hypotheses in order for the combined shape to better fit the observations. All explorations are simulated and evaluated in MATLAB. The resulting shapes are evaluated based on their Jaccard Index with the ground truth model. The algorithm leads to improvements in the Jaccard Index, but not to drastically different looking shapes.

Object-centric geometric perception aims at extracting the geometric attributes of 3D objects. These attributes include shape, pose, and motion of the target objects, which enable fine-grained object-level understanding for various tasks in graphics, computer vision, and robotics. With the growth of 3D geometry data and 3D deep learning methods, it becomes more and more likely to achieve such tasks directly using 3D input data. Among different 3D representations, a 3D point cloud is a simple, common, and memory-efficient representation that could be directly retrieved from multi-view images, depth scans, or LiDAR range images. Different challenges exist in achieving object-centric geometric perception, such as achieving a fine-grained geometric understanding of common articulated objects with multiple rigid parts, learning disentangled shape and pose representations with fewer labels, or tackling dynamic and sequential geometric input in an end-to-end fashion. Here we identify and solve these challenges from a 3D deep learning perspective by designing effective and generalizable 3D representations, architectures, and pipelines. We propose the first deep pose estimation for common articulated objects by designing a novel hierarchical invariant representation. To push the boundary of 6D pose estimation for common rigid objects, a simple yet effective self-supervised framework is designed to handle unlabeled partial segmented scans. We further contribute a novel 4D convolutional neural network called PointMotionNet to learn spatio-temporal features for 3D point cloud sequences. All these works advance the domain of object-centric geometric perception from a unique 3D deep learning perspective. / Doctor of Philosophy / 3D sensors these days are widely equipped on various mobile devices like a depth camera on iPhone, or laser LiDAR sensors on an autonomous driving vehicle. These 3D sensing techniques could help us get accurate measurements of the 3D world. For the field of machine intel- ligence, we also want to build intelligent system and algorithm to learn useful information and understand the 3D world better. We human beings have the incredible ability to sense and understand this 3D world through our visual or tactile system. For example, humans could infer the geometry structure and arrangement of furniture in a room without seeing the full room, we are able to track an 3D object no matter how its appearance, shape and scale changes, we could also predict the future motion of multiple objects based on sequential observation and complex reasoning. Here my work designs various frameworks to learn such 3D information from geometric data represented by a lot of 3D points, which achieves fine-grained geometric understanding of individual objects, and we can help machine tell the target objects' geometry, states, and dynamics. The work in this dissertation serves as building blocks towards a better understanding of this dynamic world.

In recent years, the EU has observed a decrease in the stocks of certain fish species due to unrestricted fishing. To combat the problem, many fisheries are investigating how to automatically estimate the catch size and composition using sensors onboard the vessels. Yet, measuring the size of fish in marine imagery is a difficult task. The images generally suffer from complex conditions caused by cluttered fish, motion blur and dirty sensors. In this thesis, we propose a novel method for automatic measurement of fish size that can enable measuring both visible and occluded fish. We use a Mask R-CNN to segment the visible regions of the fish, and then fill in the shape of the occluded fish using a U-Net. We train the U-Net to perform shape completion in a semi-supervised manner, by simulating occlusions on an open-source fish dataset. Different to previous shape completion work, we teach the U-Net when to fill in the shape and not by including a small portion of fully visible fish in the input training data. Our results show that our proposed method succeeds to fill in the shape of the synthetically occluded fish as well as of some of the cluttered fish in real marine imagery. We achieve an mIoU score of 93.9 % on 1 000 synthetic test images and present qualitative results on real images captured onboard a fishing vessel. The qualitative results show that the U-Net can fill in the shapes of lightly occluded fish, but struggles when the tail fin is hidden and only parts of the fish body is visible. This task is difficult even for a human, and the performance could perhaps be increased by including the fish appearance in the shape completion task. The simulation-to-reality gap could perhaps also be reduced by finetuning the U-Net on some real occlusions, which could increase the performance on the heavy occlusions in the real marine imagery.

In this thesis we explore the concepts and components which can be used as individual building blocks for producing immersive virtual reality (VR) content from a single RGB-D sensor. We identify the properties of immersive VR videos and propose a system composed of a foreground/background separator, a dynamic scene re-constructor and a shape completer. We initially explore the foreground/background separator component in the context of video summarization. More specifically, we examined how to extract trajectories of moving objects from video sequences captured with a static camera. We then present a new approach for video summarization via minimization of the spatial-temporal projections of the extracted object trajectories. New evaluation criterion are also presented for video summarization. These concepts of foreground/background separation can then be applied towards VR scene creation by extracting relative objects of interest. We present an approach for the dynamic scene re-constructor component using a single moving RGB-D sensor. By tracking the foreground objects and removing them from the input RGB-D frames we can feed the background only data into existing RGB-D SLAM systems. The result is a static 3D background model where the foreground frames are then super-imposed to produce a coherent scene with dynamic moving foreground objects. We also present a specific method for extracting moving foreground objects from a moving RGB-D camera along with an evaluation dataset with benchmarks. Lastly, the shape completer component takes in a single view depth map of an object as input and "fills in" the occluded portions to produce a complete 3D shape. We present an approach that utilizes a new data minimal representation, the additive depth map, which allows traditional 2D convolutional neural networks to accomplish the task. The additive depth map represents the amount of depth required to transform the input into the "back depth map" which would exist if there was a sensor exactly opposite of the input. We train and benchmark our approach using existing synthetic datasets and also show that it can perform shape completion on real world data without fine-tuning. Our experiments show that our data minimal representation can achieve comparable results to existing state-of-the-art 3D networks while also being able to produce higher resolution outputs.