Semantic Segmentation on Roads

Project Overview

The Semantic Segmentation on Roads project aimed to improve segmentation accuracy on road scenes by leveraging an encoder-decoder DNN structure. Various architectures, including GCN (Graph Convolutional Network), PSPNet (Pyramid Scene Parsing Network), and DUC-HDC (Dense Upsampling Convolution & Hybrid Dilated Convolution), were implemented and compared to assess their performance improvements on road scene datasets.

Objectives

• Enhanced Segmentation Accuracy: Apply and evaluate advanced segmentation models for detailed road scene analysis.
• Architecture Comparison: Examine the impact of GCN, PSPNet, and DUC-HDC on segmentation accuracy.

Highlights

Project achievements include:

• A comparative study revealing the strengths of each architecture in terms of segmentation detail, speed, and resource consumption.
• Enhanced accuracy in road lane and vehicle segmentation through DNN optimizations.
• Insights into architecture trade-offs, supporting further research in semantic segmentation.

Visual Insights

Road Scene Segmentation

Architecture used for accurate road scene analysis and segmentation.

Additional Resources

For more information, view the project’s GitHub repository