TWIST: Trains under Weather, Illumination, and Seasonal Time

This repository accompanies the paper:

TWIST: Trains under Weather, Illumination, and Seasonal Time
Momin Ali, Andre Stenger, Til Arkenberg, Laura Harms, Olaf Landsiedel
ISIoT Workshop at DCOSS 2026

TWIST is a real-world dataset for train detection designed to improve robustness of vision-based railway monitoring systems under diverse environmental conditions. Vision-based railway monitoring systems often fail in real-world deployments due to limited training data diversity. TWIST addresses this gap by providing a dataset collected across multiple seasons, capturing:

• 🌧️ Rain
• ❄️ Snow
• 🌫️ Fog
• 🌙 Low-light & night
• ☀️ Glare & daylight
• 🚄 Motion blur & varying train speeds

Check out the supplied Jupyter Notebooks to analyze the data in our TWIST GitHub repo

• Total Images: ~38,000
• Resolution: 640 × 480
• Binary Annotations: 10,000
• Detailed Annotations: 1,493 images

1. Binary Labels

   • Train / No Train

2. Detailed labels

   • Locomotive
   • Wagon
   • Freight Car
   • High-Speed Train

Annotations follow the YOLO format:

<class_id> <x_center> <y_center> <width> <height>

• All values are normalized between 0 and 1
• Compatible with YOLOv5, YOLOv8, and other frameworks

This project is licensed under the terms of the Creative Commons Attribution 4.0 International License.