Visual Bank Inc. (Minato-ku, Tokyo; CEO: Saneyuki Nagai), through its subsidiary amanaimages Inc., has launched a new dataset under its AI training data solution, Qlean Dataset: the Japanese Railway Level Crossing Image Dataset.
The dataset is designed for image-based AI development and research, including object detection, scene recognition, and multimodal AI applications in the railway and transportation infrastructure domain.

This dataset consists of images of railway level crossings located throughout Japan, captured from a variety of angles, distances, and surrounding environments.
The images include characteristic crossing facilities such as barriers, warning signals, and crossing signs, and record different operational states of level crossings, including before, during, and after train passage.

The dataset covers a wide range of locations, from level crossings in densely populated urban residential areas to those along regional rail lines, coastal areas, and suburban environments.
By reflecting the diverse placement conditions and surrounding contexts of Japan’s railway infrastructure, the dataset enables the use of training data that is not biased toward a single environment, supporting learning of real-world visual appearances and structural characteristics.

The Japanese Railway Level Crossing Image Dataset is provided as part of Qlean Dataset’s original data lineup, AI Data Recipe, and is suitable for use from academic research to AI development intended for commercial deployment, with careful consideration of legal and usage risks.
Visual Bank will continue to develop and provide training data that reflects real-world environments, contributing to the foundation of AI development across transportation, infrastructure, and related fields.

Dataset Overview: Japanese Railway Level Crossing Image Dataset

Example Use Cases for the Japanese Railway Level Crossing Image Dataset

Research Applications

• Object and Structural Understanding in Railway Crossing Scenes
  Used for training and evaluating object detection and structural understanding models in complex outdoor traffic environments.

• Robustness and Generalization Testing for Vision Models
  Supports evaluation of model performance across diverse locations and environmental conditions within railway infrastructure scenes.

• Multimodal Traffic Scene Understanding
  Applicable to evaluating traffic scene comprehension when combined with text or time-series data in multimodal AI models.

Industrial Applications

• Safety Monitoring AI for Railway Infrastructure
  Used as training and validation data for AI systems that recognize level crossing equipment and surrounding environments.

• Pretraining and Fine-Tuning for Traffic-Oriented Foundation Models
  Provides real-world railway crossing imagery for vision and multimodal foundation model training.

• Evaluation Data for Edge AI and Fixed-Camera Systems
  Suitable for benchmarking object detection and scene recognition models under surveillance-style camera conditions.

Additional Practical Use

• AI Education and Proof-of-Concept in Transportation Domains
  Supports education and demonstration projects focused on image recognition and scene understanding using real infrastructure data.