This repository contains the experimental design of my master's thesis, broken down into these sections:
I used the KITTI and AVX data sets for performance evaluation, with the 'Train' subsets for model training, and the 'Test' subsets for validation and evaluation. For more details, see the readme.
I used the PointPillars Network for 3D object detection, with the OpenPCDet tool. The model targets 'Car', 'Pedestrian', and 'Cyclist' classes. Training parameters and techniques are also discussed. For more details, see the readme.
The KITTI Evaluation Metrics are used for network assessment, which includes the 3D Intersection over Union (IoU), Precision-Recall (PR) curve, Average Precision (AP), and Average Orientation Similarity (AOS) metric. For more details, see the readme.
I examined the applicability of KITTI difficulty levels to synthetic data sets in object detection context. For more details, see the readme.
A collection of experiments are included, which focus on neural network performance when trained on both real and synthetic data sets. For more details, see the readme.