DOI: https://doi.org/10.61957/joerai-20250206

Title: Advanced Artificial Intelligence

Author:TAFERE GEDION LEUL

Abstract

This report details the development of a semantic segmentation model using the DeepLabV3+ architecture with a Mobile Net backbone, trained on the Pascal VOC2012 dataset. The objective was to create a computationally efficient model capable of generating accurate pixel-wise segmentation masks (single-channel grayscale) for real-world applications. The project encompassed data preprocessing, model construction, training (utilizing Tensor Board for visualization), and inference. Key learning outcomes included managing segmentation data, designing/tuning neural networks, and optimizing training processes.

Keywords:Semantic Segmentation; DeepLabV3+; Mobile Net; Pascal VOC2012