3D Object Detection in Context

Document Type

Dissertation

Abstract

This work presents a second contribution attempting to push the performance of the contemporary state-of-the-art 3D object detector, RBGNet, by introducing self-attention on multiple levels. Inspired by self-attention is introduced at; (1) the point patch level to capture correlations between parts, or geometries, of objects, (2) the object candidate level to capture relationships between objects in the scene, and (3) the scene level to capture contextual cues. Through a series of experiments, the introduced self-attention modules prove to have a positive effect on the performance of the RBGNet baseline.

First Page

i

Last Page

43

Publication Date

12-1-2022

Comments

Thesis submitted to the Deanship of Graduate and Postdoctoral Studies

In partial fulfillment of the requirements for the M.Sc degree in Computer Vision

Advisors: Dr. Rao Anwer, Dr. Muhammad Haris

Online access for MBZUAI patrons

Share

COinS