Overview
The Rail-DB_NTA (Rail-DB Navigable Track Area) is a collection of real-world train track images captured under three different lighting conditions, the images are selected from the “fixed_day”, “fixed_night”, “fixed_rain”, and “fixed_rain_2” subsets of the Rail-DB dataset present by Xinpeng Li. The images are then re-annotated by a new novel strategy that jointly labels track lines and navigable areas. Given the unique challenges of railway track--such as turnouts and intricate curvature variations--this dataset aims to support research in autonomous railway systems, track detection algorithms, and navigable area identification solutions.

Key Features
	Challenging conditions: Includes ane changes through turnouts, and track line intersections in the distance and occlusions.
	Diverse environments: Day, night and rainy recordings, different track segments, and varying types of railway.
	New novel annotation stratege: The tracklines are divided into five categories and implicitly annotates the current track the train is traveling on. This design enables the network to detect different categories of track lines and segment the navigable area more accurately.
	Balance of each sample: Several samples are manually augmented to balance the data and prevent certain track types from being hard to detect.

Data Stucture
This folder contains the dataset required for the experiments, and the results generated by the network we proposed.
1、Dataset: the dataset is located in the “data” folder, which contains the original images, and the corresponding mask images.
	The dataset for each major scenario are spilted into “train” and “valid” categories, and each contains its image and mask subfolders.
	The folders with the" _transform "suffix contain the transformed images and their corresponding masks.
2、Network Output:The output of the network is stored in the result folder, which includes:
	Network Output: The results from the optimized UfastNet we proposed, which detect all tracks in the image. These results are stored in the “network_output” folder.
	Post-Processed Output: The navigable area results obtained after applying our post-processing algorithm, stored in the “after_post_operation” folder.

Usage Instructions
1、Downloading the Dataset
	The dataset is available as a ZIP file containing original images, json files and the corresponding mask images.
	Extract the ZIP file to access.
2、Understanding the Data
	Each image is named based on its lighting conditions and the sequence of collection (e.g., fixed_day_000001.jpeg).
	The corresponding json file and mask image has the same name as the original image.
3、Processing the Data
	Recommended tools: OpenCV, TensorFlow, PyTorch for track detection.
	Utilize deep learning models for segmentation and object detection.
4、Potential Applications
	Track detection and occlusion warning systems.
	Autonomous railway navigation research.
	Road safety and accident prediction models.