IFD: An RGB–SSM Dataset for Fine-Grained Insulator Anomaly Detection

<p>The IFD dataset is released to support research on fine-grained insulatoranomaly detection under complex outdoor inspection conditions. The datasetcontains paired RGB images and RGB-derived Surface Structural Modality(SSM) tensors, together with object annotations and fixed data splits.</p> <p>Version 1.0 includes 2,843 RGB–SSM image pairs, comprising 2,165 trainingsamples, 391 validation samples, and 287 test samples. The releasedannotations follow the YOLO object-detection format.</p> <p>The released files include:</p> <ul><li>Original RGB images</li><li>YOLO-format object annotations</li><li>Fixed training, validation, and test split files</li><li>Generated six-channel SSM tensors</li><li>Class-definition and dataset-statistics files</li><li>File-integrity checksums</li></ul> <p>Each SSM tensor is derived from the corresponding RGB image and organizessix complementary structural descriptors: luminance, gradient magnitude,gradient orientation, Laplacian response, local roughness, and localhigh-frequency residual. RGB images and SSM tensors share the same filestem and are spatially aligned.</p> <p>The associated DBRM-Net source code, training and evaluation configurations,and SSM generation scripts are maintained in the accompanying coderepository. Users may either use the released SSM tensors directly orregenerate them from the RGB images using the provided scripts.</p> <p>Please refer to the README file for the directory structure, classdefinitions, split protocol, file correspondence, and reproductioninstructions.</p>