Neural differential geometry for 3D object detection
This project explores advanced techniques and current trends in 2D and 3D object detection, focusing on overcoming challenges in the field through the development of a multi-modal architecture that leverages both camera and LiDAR data. Our approach involves a critical analysis of the components of a...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/178656 |
| Summary: | This project explores advanced techniques and current trends in 2D and 3D object detection, focusing on overcoming challenges in the field through the development of a multi-modal architecture that leverages both camera and LiDAR data. Our approach involves a critical analysis of the components of an existing model and the integration of innovative neural Ordinary Differential Equation (ODE) blocks into the 3D Region Proposal Network (RPN) backbone. We introduce two novel ODE-based models aimed at enhancing the dynamic and deep feature extraction capabilities of traditional object detection systems. Extensive training and testing have demonstrated that our ODE-enhanced model outperforms the original in terms of accuracy and efficiency. These findings suggest significant potential for the application of neural ODE methodologies in advancing 3D object detection technologies. |
|---|