Reasoning about 3D shape of objects is important for successful computer visionapplications in robotics, 3D rendering and modeling. In this thesis, we address twoproblems { First, given an image, we generate 3D shape of the foreground object thatappears in the image. Second, we predict the class label of the input...
Soft robots are designed to utilize their compliance and contortionistic abilities to both interact safely with their environment and move through it in ways a rigid robot cannot. To more completely achieve this, the robot should be made of as many soft components as possible. Here we present a completely...
Currently, a popular approach to image classification uses the deep Transformer architecture. In a Transformer, the attention mechanism enables the model to learn efficiently with fewer computational resources than the convolutional neural networks (CNNs). In this thesis, we study the sparse attention mechanism widely used in the Transformers developed specifically...
Many object recognition applications require detecting and responding to objects drawn from a different distribution from that of the training data. This task is referred to as out-of-distribution (OOD) detection, and it is often formulated as an outlier detection problem
wherein the probability distribution of the known data P(X) is...
Assessing AI systems is difficult. Humans rely on AI systems in increasing ways, both visible and invisible, meaning a variety of stakeholders need a variety of assessment tools (e.g., a professional auditor, a developer, and an end user all have different needs). We posit that it is possible to provide...
The local presence or absence of individual botanical species can be predicted with high accuracy by a simple feed-forward neural network, using only local climate data to make inference. This study proposes a framework for learning these predictive models, demonstrates highly accurate predictions for species with a sufficiently large area...
Recent studies have shown that novel continuous dropout methods can be viewed as a Bayesian interpretation of model parameters, though most such studies have shown results using normal distributions. As the posterior distributions over neural network nodes and parameters are intractable, given that they are a result of artificial construction...
This report presents an efficient method for semi-supervised video object segmentation – the problem of identifying foreground pixels occupied by a target object. The target is specified by the ground-truth mask in the first video frame. While the state of the art achieves a segmentation accuracy greater than 80%, it...
This thesis considers the problem of training convolutional neural networks for online visual tracking. A major challenge for single object visual tracking is that most training sets with frame-level track annotations are quite small, due to the prohibitive cost of manual annotation. Current training approaches either supplement the annotations with...
This thesis is about visual relationship detection. This is an important task in computer vision. The goal is to detect all visual relationships in a given image between objects. This thesis presents a new approach to this problem. Our approach does not use an object detector as a common pre-processing...