Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Racing Dataset and Baseline Model for Track Detection in Autonomous Racing

Shreya Ghosh, Yi-Huan Chen, Ching-Hsiang Huang, Abu Shafin Mohammad Mahdee Jameel, Chien Chou Ho, Aly El Gamal, Samuel Labi

TL;DR

RoRaTrack delivers the first open, real-world multi-camera racing dataset specifically designed for track detection on road courses, addressing challenges unique to autonomous racing such as high-speed blur and absent lane markings. The authors propose RaceGAN, a GAN-based baseline with a WeBACNN-inspired generator and pixel-wise discriminator, trained with a domain-transfer loss to adapt to racing data and enhanced by post-processing. Quantitative results show RaceGAN achieving state-of-the-art performance on RoRaTrack (e.g., $mIoU=0.8691$, $Accuracy=0.9580$, $F1=0.8738$) while maintaining real-time inference, outperforming seven traffic-lane baselines and one track method. The work provides an open dataset and code, enabling future development of racing-specific track-detection methods with practical impact for autonomous racing systems.

Abstract

A significant challenge in racing-related research is the lack of publicly available datasets containing raw images with corresponding annotations for the downstream task. In this paper, we introduce RoRaTrack, a novel dataset that contains annotated multi-camera image data from racing scenarios for track detection. The data is collected on a Dallara AV-21 at a racing circuit in Indiana, in collaboration with the Indy Autonomous Challenge (IAC). RoRaTrack addresses common problems such as blurriness due to high speed, color inversion from the camera, and absence of lane markings on the track. Consequently, we propose RaceGAN, a baseline model based on a Generative Adversarial Network (GAN) that effectively addresses these challenges. The proposed model demonstrates superior performance compared to current state-of-the-art machine learning models in track detection. The dataset and code for this work are available at https://github.com/ghosh64/RaceGAN.

A Racing Dataset and Baseline Model for Track Detection in Autonomous Racing

TL;DR

RoRaTrack delivers the first open, real-world multi-camera racing dataset specifically designed for track detection on road courses, addressing challenges unique to autonomous racing such as high-speed blur and absent lane markings. The authors propose RaceGAN, a GAN-based baseline with a WeBACNN-inspired generator and pixel-wise discriminator, trained with a domain-transfer loss to adapt to racing data and enhanced by post-processing. Quantitative results show RaceGAN achieving state-of-the-art performance on RoRaTrack (e.g., , , ) while maintaining real-time inference, outperforming seven traffic-lane baselines and one track method. The work provides an open dataset and code, enabling future development of racing-specific track-detection methods with practical impact for autonomous racing systems.

Abstract

A significant challenge in racing-related research is the lack of publicly available datasets containing raw images with corresponding annotations for the downstream task. In this paper, we introduce RoRaTrack, a novel dataset that contains annotated multi-camera image data from racing scenarios for track detection. The data is collected on a Dallara AV-21 at a racing circuit in Indiana, in collaboration with the Indy Autonomous Challenge (IAC). RoRaTrack addresses common problems such as blurriness due to high speed, color inversion from the camera, and absence of lane markings on the track. Consequently, we propose RaceGAN, a baseline model based on a Generative Adversarial Network (GAN) that effectively addresses these challenges. The proposed model demonstrates superior performance compared to current state-of-the-art machine learning models in track detection. The dataset and code for this work are available at https://github.com/ghosh64/RaceGAN.

Paper Structure

This paper contains 20 sections, 3 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Track Detection Methods
  4. Existing Datasets
  5. RoRaTrack Dataset
  6. Data Collection and Split
  7. Dataset Characteristics
  8. Annotation
  9. Proposed Method
  10. Generator
  11. Discriminator
  12. GAN Training
  13. Post Processing
  14. Results
  15. Evaluation Metrics
  16. ...and 5 more sections

Figures (4)

  • Figure 1: Images depicting normal and challenging road scenarios—such as dazzle light, color imbalance, curved roads, and blurriness.
  • Figure 2: The architecture of the proposed RaceGAN model is depicted, highlighting the detailed structure of both the generator and discriminator blocks. Additionally, the training flow for the discriminator and the generator is illustrated separately.
  • Figure 3: Architecture of the WeBACNN block ghosh2024weighted.
  • Figure 4: Representative images illustrating the track prediction process: (a) The original input image, (b) The predicted lane mask, and (c) The superimposed image displaying the predicted lane (highlighted in red) overlaid on the original image.