Deep neural networks (DNNs), which are based loosely on the ventral stream pathway of the primate visual system, are good models of the visual system …
Crowdsourced training data has become a mainstay in computer vision. Some of the most significant discoveries of the last few years were made possible by …
The problem of detecting changes in images taken by a stationary camera has been well studied and many algorithms now exist which perform robustly in …
Human communication is naturally multimodal. On the web, images frequently appear alongside text: for example, product images and descriptions on shopping websites, or social media …
This thesis develops novel algorithms to automate video analysis for fixed monocular surveillance cameras. Specifically, change detection algorithms are proposed to identify spatio-temporal locations in …
We are currently working on adding more autonomy to quadrotor drones so that they can execute a desired GPS coordinate path while avoiding unexpected obstacles. …
Advancements in machine learning techniques have encouraged scholars to focus on convolutional neural network (CNN) based solutions for object detection and pose estimation tasks. Most …
Vehicle surveillance is the task of measuring moving road vehicles to automatically obtain information about vehicle shape, appearance, identity, path of motion, and, ultimately, driver …
Researchers at Brown University have developed novel algorithms for low level object recognition. While they are remarkably accurate, their methods are computationally expensive and as …
Recent advances in deep learning have enabled new generative models for 3D objects, in particular manufactured shapes (e.g. chairs, tables, airplanes, cars, ...). The most …
Obtaining accurate markerless 3D pose estimations of subjects is an important problem in neuroscience and other fields. Recent work has employed CNNs very successfully to …
Many applications in Computer Vision and Machine Learning entail learning from partially annotated data. A popular family of models that can capture unobserved variables in …
This thesis develops the multiple view geometry of arbitrary, piecewise differentiable curves using differential geometry, and the beginnings of a theory on general surfaces. These …
We propose a new paradigm for vision-based human motion capture. This paradigm extends the traditional capture of poses by providing guarantees of physical plausibility for …
In recent years, object recognition on image has achieved great success. Well-trained CNNs can successfully detect objects in a single image with high accuracy. However, …
This thesis is concerned with estimating the time-varying 3-d surface geometry and surface appearance of dynamic scenes from images. An accurate reconstruction of a 4-d …
Face recognition is a widely studied area in computer vision. Several deep-learning-based face recognition models can achieve high accuracies on large datasets and general classification …