A5040003361- 3D Shape Modeling and Analysis
- 3D Surveying and Cultural Heritage
- Advanced Vision and Imaging
- Computer Graphics and Visualization Techniques
- Human Pose and Action Recognition
- Remote Sensing and LiDAR Applications
- Optical measurement and interference techniques
- Advanced Neural Network Applications
- Religion, Society, and Development
- Image Processing and 3D Reconstruction
- Nonprofit Sector and Volunteering
- Medical Image Segmentation Techniques
- Satellite Image Processing and Photogrammetry
- Advanced Numerical Analysis Techniques
- Robotics and Sensor-Based Localization
- Medical Imaging and Analysis
- Anatomy and Medical Technology
- Religion and Society Interactions
Rice University
2024
The University of Queensland
2019-2024
Queensland University of Technology
2020
Implicit shape representations, such as Level Sets, provide a very elegant formulation for performing computations involving curves and surfaces. However, including implicit representations into canonical Neural Network formulations is far from straightforward. This has consequently restricted existing approaches to inference, significantly less effective perhaps most commonly voxels occupancy maps or sparse point clouds. To overcome this limitation we propose novel that permits the use of...
Deep learning applied to the reconstruction of 3D shapes has seen growing interest. A popular approach and generation in recent years been CNN encoder-decoder model usually voxel space. However, this often scales very poorly with resolution limiting effectiveness these models. Several sophisticated alternatives for decoding have proposed typically relying on complex deep architectures decoder model. In work, we show that additional complexity is not necessary, can actually obtain high...
In conventional formulations of multilayer feedforward neural networks, the individual layers are customarily defined by explicit functions. this paper we demonstrate that defining in a network \emph{implicitly} provide much richer representations over standard one, consequently enabling vastly broader class end-to-end trainable architectures. We present general framework implicitly layers, where theoretical analysis such can be addressed through implicit function theorem. also show how...
The impressive performance of deep convolutional neural networks in single-view 3D reconstruction suggests that these models perform non-trivial reasoning about the structure output space. Recent work has challenged this belief, showing that, on standard benchmarks, complex encoder-decoder architectures similarly to nearest-neighbor baselines or simple linear decoder exploit large amounts per-category data. However, building collections shapes for supervised training is a laborious process;...
3D building reconstruction from imaging data is an important task for many applications ranging urban planning to reconnaissance. Modern Novel View synthesis (NVS) methods like NeRF and Gaussian Splatting offer powerful techniques developing models natural 2D imagery in unsupervised fashion. These algorithms generally require input training views surrounding the scene of interest, which, case large buildings, typically not available across all camera elevations. In particular, most readily...
In this paper, we analyze the viewpoint stability of foundational models - specifically, their sensitivity to changes in viewpoint- and define instability as significant feature variations resulting from minor viewing angle, leading generalization gaps 3D reasoning tasks. We investigate nine models, focusing on responses changes, including often-overlooked accidental viewpoints where specific camera orientations obscure an object's true structure. Our methodology enables recognizing...
The impressive performance of deep convolutional neural networks in single-view 3D reconstruction suggests that these models perform non-trivial reasoning about the structure output space. Recent work has challenged this belief, showing that, on standard benchmarks, complex encoder-decoder architectures similarly to nearest-neighbor baselines or simple linear decoder exploit large amounts per-category data. However, building collections shapes for supervised training is a laborious process;...