Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated Road Distress Detection Using Vision Transformersand Generative Adversarial Networks

Cesar Portocarrero Rodriguez, Laura Vandeweyen, Yosuke Yamamoto

TL;DR

This project explores the use of state-of-the-art CV techniques for road distress segmentation with results showing that GAN-generated data improves model performance and that MaskFormer outperforms the CNN model in two metrics: mAP50 and IoU.

Abstract

The American Society of Civil Engineers has graded Americas infrastructure condition as a C, with the road system receiving a dismal D. Roads are vital to regional economic viability, yet their management, maintenance, and repair processes remain inefficient, relying on outdated manual or laser-based inspection methods that are both costly and time-consuming. With the increasing availability of real-time visual data from autonomous vehicles, there is an opportunity to apply computer vision (CV) methods for advanced road monitoring, providing insights to guide infrastructure rehabilitation efforts. This project explores the use of state-of-the-art CV techniques for road distress segmentation. It begins by evaluating synthetic data generated with Generative Adversarial Networks (GANs) to assess its usefulness for model training. The study then applies Convolutional Neural Networks (CNNs) for road distress segmentation and subsequently examines the transformer-based model MaskFormer. Results show that GAN-generated data improves model performance and that MaskFormer outperforms the CNN model in two metrics: mAP50 and IoU.

Automated Road Distress Detection Using Vision Transformersand Generative Adversarial Networks

TL;DR

This project explores the use of state-of-the-art CV techniques for road distress segmentation with results showing that GAN-generated data improves model performance and that MaskFormer outperforms the CNN model in two metrics: mAP50 and IoU.

Abstract

The American Society of Civil Engineers has graded Americas infrastructure condition as a C, with the road system receiving a dismal D. Roads are vital to regional economic viability, yet their management, maintenance, and repair processes remain inefficient, relying on outdated manual or laser-based inspection methods that are both costly and time-consuming. With the increasing availability of real-time visual data from autonomous vehicles, there is an opportunity to apply computer vision (CV) methods for advanced road monitoring, providing insights to guide infrastructure rehabilitation efforts. This project explores the use of state-of-the-art CV techniques for road distress segmentation. It begins by evaluating synthetic data generated with Generative Adversarial Networks (GANs) to assess its usefulness for model training. The study then applies Convolutional Neural Networks (CNNs) for road distress segmentation and subsequently examines the transformer-based model MaskFormer. Results show that GAN-generated data improves model performance and that MaskFormer outperforms the CNN model in two metrics: mAP50 and IoU.

Paper Structure

This paper contains 22 sections, 2 equations, 15 figures.

Table of Contents

  1. Introduction
  2. Related work
  3. Generative adversarial networks
  4. Vision transformers
  5. Convolutional neural networks
  6. Materials and Methods
  7. Synthetic data generation using GANs
  8. Building the Generator and Discriminator Models
  9. Defining Loss Functions and Optimizer
  10. YOLOv8 implementation
  11. YOLOv8 backbone
  12. YOLOv8 anchor-free detection head
  13. YOLOv8 loss function
  14. MaskFormer implementation
  15. Mask classification formulation
  16. ...and 7 more sections

Figures (15)

  • Figure S1: Architecture selected for the generator and discriminator models.
  • Figure S2: Diagram of YOLOv8 backbone architecture. Github_YOLOarch
  • Figure S3: Detection head for Yolov8 Github_YOLOarch
  • Figure S4: MaskFormer architecture and its three modules: pixel-level, transformer and segmentation. MaskFormer
  • Figure S5: Example image from the NDTI dataset. (a) Raw image. (b) Annotated images with cracks (red) and damaged guardrail (blue).
  • ...and 10 more figures