Current Structure-from-Motion (SfM) methods typically follow a two-stage pipeline, combining learned or geometric pairwise reasoning with a subsequent global optimization step. In contrast, we propose a data-driven multi-view reasoning approach that directly infers 3D scene geometry and camera poses from multi-view images. Our framework, DiffusionSfM, parameterizes scene geometry and cameras as pixel-wise ray origins and endpoints in a global frame and employs a transformer-based denoising diffusion model to predict them from multi-view inputs. To address practical challenges in training diffusion models with missing data and unbounded scene coordinates, we introduce specialized mechanisms that ensure robust learning. We empirically validate DiffusionSfM on both synthetic and real datasets, demonstrating that it outperforms classical and learning-based approaches while naturally modeling uncertainty.

We address the task of estimating 6D camera poses from sparse-view image sets (2-8 images). This task is a vital pre-processing stage for nearly all contemporary (neural) reconstruction algorithms but remains challenging given sparse views, especially for objects with visual symmetries and texture-less surfaces. We build on the recent RelPose framework which learns a network that infers distributions over relative rotations over image pairs. We extend this approach in two key ways; first, we use attentional transformer layers to process multiple images jointly, since additional views of an object may resolve ambiguous symmetries in any given image pair (such as the handle of a mug that becomes visible in a third view). Second, we augment this network to also report camera translations by defining an appropriate coordinate system that decouples the ambiguity in rotation estimation from translation prediction. Our final system results in large improvements in 6D pose prediction over prior art on both seen and unseen object categories and also enables pose estimation and 3D reconstruction for in-the-wild objects.

Estimating camera poses is a fundamental task for 3D reconstruction and remains challenging given sparsely sampled views (<10). In contrast to existing approaches that pursue top-down prediction of global parametrizations of camera extrinsics, we propose a distributed representation of camera pose that treats a camera as a bundle of rays. This representation allows for a tight coupling with spatial image features improving pose precision. We observe that this representation is naturally suited for set-level transformers and develop a regression-based approach that maps image patches to corresponding rays. To capture the inherent uncertainties in sparse-view pose inference, we adapt this approach to learn a denoising diffusion model which allows us to sample plausible modes while improving performance. Our proposed methods, both regression- and diffusion-based, demonstrate state-of-the-art performance on camera pose estimation on CO3D while generalizing to unseen object categories and in-the-wild captures.

Most of the climatological studies derived from reanalysis datasets to-date have been presence-based rather than event-based. We have gathered event-based climatological statistics from thirty-seven-plus (37+) years of blizzard-like snowstorms, identified and individually tracked, using hourly high-resolution datasets from the NASA's Modern Era Retrospective-analysis for Research and Applications (MERRA) data collection. We have not only extracted summary statistics for all storms, such as cumulative-probability density functions (CDFs, in percentiles) of storm duration and cumulative area coverage, but also per-event statistics for each storm, e.g. beginning/ending times, hourly locations, hourly mean snowfall intensities, and snowfall intensity probability distribution. In addition, we have constructed a web portal where users can view the hourly locations of each snowstorm annotated with conditions of the storm at that hour. Users with accounts on the portal can discover coincident data granules of relevant satellite remote-sensing observations by querying NASA metadata repository, i.e. EOS Clearing House (ECHO) or upcoming Common Metadata Repository (CMR) and create/share individualized virtual collections apposite to their research.

We address the problem of 3D medical volume reconstruction using Web services. The use of proposed Web services is motivated by the fact that the problem of 3D medical volume reconstruction requires significant computer resources and human expertise in medical and computer science areas. Web services were implemented as an additional layer to a dataflow framework called data to knowledge. In the collaboration between UIC and NCSA, pre-processed input images at NCSA were made accessible to medical collaborators for registration. Every time medical collaborators inspected images and selected corresponding features for registration, the Web server at NCSA was contacted and the registration processing query was executed using the image to knowledge library of registration methods. Co-registered frames were returned for verification by medical collaborators in a new window. This paper presents 3D volume reconstruction problem requirements, architecture of the developed prototype system and the tradeoffs of our system design.

Ensemble classifier methods have been developed to learn from streaming data, and adapt to concept drift. One strategy employed to adapt to concept drift is to rank the classifiers in the ensemble based on their performance on the most recent samples. However, this strategy is problematic when the samples are coming from different portions of the sample space during different times. Data streams with unevenly distributed samples can occur in a wide variety of applications including detection of seasonally occurring weather phenomena, processing of video streams and others. In applications such as these, the uneven distribution of the data in the sample space must be accounted for in order to derive classifiers with good performance. In order to address this requirement, the Coverage Based Ensemble Algorithm (CBEA) was developed. CBEA is an incremental anytime learning algorithm designed to work on streaming data where the samples are very unevenly distributed. The CBEA algorithm keeps information about the range of data used to train each classifier, and keeps classifiers based on two factors: their coverage of the range of possible values, and their age. The CBEA algorithm is described in detail and experimental results demonstrating its utility are presented.