A5014151797- Robotics and Sensor-Based Localization
- Advanced Neural Network Applications
- 3D Shape Modeling and Analysis
- Remote Sensing and LiDAR Applications
- Video Surveillance and Tracking Methods
- Stellar, planetary, and galactic studies
- Rough Sets and Fuzzy Logic
- Machine Learning and Data Classification
- 3D Surveying and Cultural Heritage
- Astronomy and Astrophysical Research
- Anomaly Detection Techniques and Applications
- Natural Language Processing Techniques
- Adversarial Robustness in Machine Learning
- Data Mining Algorithms and Applications
- Galaxies: Formation, Evolution, Phenomena
- Imbalanced Data Classification Techniques
California Institute of Technology
2020-2023
By identifying four important components of existing LiDAR-camera 3D object detection methods (LiDAR and camera candidates, transformation, fusion outputs), we observe that all either find dense candidates or yield representations scenes. However, given objects occupy only a small part scene, finding generating is noisy inefficient. We propose SparseFusion, novel multi-sensor method exclusively uses sparse representations. Specifically, SparseFusion utilizes the outputs parallel detectors in...
By identifying four important components of existing LiDAR-camera 3D object detection methods (LiDAR and camera candidates, transformation, fusion outputs), we observe that all either find dense candidates or yield representations scenes. However, given objects occupy only a small part scene, finding generating is noisy inefficient. We propose SparseFusion, novel multi-sensor method exclusively uses sparse representations. Specifically, SparseFusion utilizes the outputs parallel detectors in...
We propose UnCLe, a standardized benchmark for Unsupervised Continual Learning of multimodal depth estimation task: Depth completion aims to infer dense map from pair synchronized RGB image and sparse map. models under the practical scenario unsupervised learning over continuous streams data. Existing methods are typically trained on static, or stationary, dataset. However, when adapting novel non-stationary distributions, they "catastrophically forget" previously learned information. UnCLe...
This paper explores the potential of leveraging language priors learned by text-to-image diffusion models to address ambiguity and visual nuisance in monocular depth estimation. Particularly, traditional estimation suffers from inherent due absence stereo or multi-view cues, lack robustness vision. We argue that prior can enhance geometric aligned with description, which is during pre-training. To generate images reflect text properly, model must comprehend size shape specified objects,...
Current LiDAR odometry, mapping and localization methods leverage point-wise representations of 3D scenes achieve high accuracy in autonomous driving tasks. However, the space-inefficiency that use limits their development usage practical applications. In particular, scan-submap matching global map representation are restricted by inefficiency nearest neighbor searching (NNS) for large-volume point clouds. To improve space-time efficiency, we propose a novel method describing using quadric...
We propose an optimization of Dr. Ross Quin-lan's C4.5 decision tree algorithm, used for data mining and classification. will show that by discretizing binning a set's continuous attributes into four groups using our novel technique called MSD-Splitting, we can significantly improve both the algorithm's accuracy efficiency, especially when applied to large sets. standard algorithm optimized two sets obtained from UC Irvine's Machine Learning Repository: Census Income Heart Disease. In...
Current LiDAR odometry, mapping and localization methods leverage point-wise representations of 3D scenes achieve high accuracy in autonomous driving tasks. However, the space-inefficiency that use limits their development usage practical applications. In particular, scan-submap matching global map representation are restricted by inefficiency nearest neighbor searching (NNS) for large-volume point clouds. To improve space-time efficiency, we propose a novel method describing using quadric...