SCENE UNDERSTANDING IN CHALLENGING SCENARIOS

PhD Thesis Proposal Defence


Title: "SCENE UNDERSTANDING IN CHALLENGING SCENARIOS"

by

Mr. Tuan Anh VU


Abstract:

In recent years, computer vision and graphics fields have witnessed significant
progress with the emergence of novel techniques and architectures to address
complex challenges. This proposal presents novel methodologies and advancements
in three key areas: transparent object segmentation, 4D (dynamic) point cloud
reconstruction, and test-time augmentation for 3D deep learning (indoor,
outdoor, autonomous driving).

The first contribution of this proposal addresses the semantic scene
understanding for transparent objects. While glass is prevalent in modern
applications, it is often treated similarly to opaque objects in scene
understanding tasks. To overcome this limitation, we propose a pyramidal
transformer encoder-decoder architecture with two novel object cues: Boundary
Feature Aware and Reflection Region Aware module.

The second part of this proposal focuses on object reconstruction from 4D
dynamic point clouds. We propose RFNet-4D++ architecture, which jointly
reconstructs objects and their motion flows from 4D point clouds. Our approach
achieves improved overall performance by leveraging both spatial and temporal
features from a sequence of point clouds. We introduce a temporal vector field
learning module that uses unsupervised learning for flow estimation combined
with supervised learning of spatial structures for object reconstruction.

Lastly, we explore using test-time augmentation for 3D point cloud learning.
When 3D shapes are sparsely represented with low point density, downstream task
performance tends to drop significantly. We leverage implicit representation
and point cloud upsampling techniques to systematically augment point cloud
data by sampling points from the reconstructed results and using them as
test-time augmented data.


Date:                   Thursday, 24 August 2023

Time:                   2:00pm - 4:00pm

Venue:                  Room 3494
                        lifts 25/26

Committee Members:      Dr. Sai-Kit Yeung (Supervisor)
                        Prof. Chi-Keung Tang (Chairperson)
                        Dr. Qifeng Chen
                        Prof. Pedro Sander
                        Dr. Rob Scharff (ISD)


**** ALL are Welcome ****