Abstract:
The process of identifying and localizing various items, usually in pictures that
represent objects found in daily life, is called object detection. Object detection
identifies each object as belonging to a specific class and creates a bounding box
around it. In this thesis we focus our study in indoor datasets. The purpose of the thesis
is to evaluate different methods of object detection in indoor datasets. We also aim to
compare these results with each other, in order to try and find the best methods for the
selected datasets.
Overall, these results highlight how crucial it is to carefully evaluate model
architectures, preprocessing methods, and dataset properties in order to fully utilize
deep learning for 3D applications. Subsequent investigations may examine techniques
to mitigate class disparities and improve model resilience in a variety of object
categories and shapes.
