Semantic Segmentation using Efficient Spatial Pyramid Network

Overview

Semantic segmentation plays a critical role in computer vision applications, enabling pixel-level understanding of images. The above model leverages an efficient spatial pyramid of dilated convolutions to address the challenges in this domain, providing superior performance with reduced computational complexity.

Key Features

• Efficient Spatial Pyramid: The network utilizes a spatial pyramid of dilated convolutions to capture multi-scale context information effectively. This architecture improves the accuracy and robustness of semantic segmentation models.
  

Structure of this repository

This repository is organized as:

• training This directory contains the source code for training the ESPNet-C and ESPNet models.
• testing This directory contains the source code for evaluating the model on RGB Images.

Results

The ESPNet-C model achieves mIoU of 70.0% on Cityscapes dataset. The ESPNet model achieves mIoU of 70.0% on Cityscapes dataset.

Output from Dataset 1

Output from Dataset 2

Output from Dataset 3