The Railway Fastener Defect Dataset

The Railway Fastener Defect Dataset (RFDD) is a metrology-based, high-quality benchmark comprising 1,350 full-scene images (2048 × 2021 pixels) captured by operational high-speed railway inspection vehicles. Unlike conventional datasets that rely on cropped single-object images, RFDD preserves the original multi-fastener layout and strictly defines its six categories—Normal, Missing, Inverted, Displaced, Deformed, and Fractured—based on measurable engineering specifications and maintenance tolerances rather than subjective visual patterns. To ensure engineering fidelity, the dataset is constructed under a rigorous framework of six visual consistency criteria (Geometric, Optical, Boundary Gradient, Noise, Texture, and Occlusion), providing high-precision, instance-level semantic annotations that support detection, localization, and segmentation tasks. By maintaining metrology-grade quality and realistic large-field-of-view imaging conditions, RFDD facilitates the objective, quantitative assessment of both automated detection algorithms and hardware perception systems in field-level applications.