C-AF-GAF Dataset

C-AF-GAF dataset, a collection of 115,010 Gramian Angular Field (GAF) images designed for developing and benchmarking machine learning models for Atrial Fibrillation (AF) detection. The dataset is composed of three distinct sets of GAF images derived from publicly available ECG recordings:Primary Dataset: Derived from 1,436 recordings from the 4th China Physiological Signal Challenge 2021. This set is pre-split into training, validation, and testing partitions.Secondary Dataset 1 (MIMIC): Derived from the MIMIC PERform AF Dataset, intended for evaluating model generalization.Secondary Dataset 2 (MIT-BIH): Derived from the MIT-BIH Atrial Fibrillation Database, also for generalization testing.All images were generated by segmenting the source ECG signals into 30-second windows. Each segment was then transformed into two corresponding 224x224 pixel GAF image representations: a Gramian Angular Summation Field (GASF) and a Gramian Angular Difference Field (GADF).The data is organized into the three main folders shown in the image. Each of these folders contains two subfolders: GASF and GADF. The primary dataset is further divided into train, test, and val directories. The final images are stored in uniquely named subfolders, with a naming convention that includes the original sample ID, segment number, and class label (e.g., sample1_segment1_label0.jpeg).