ECT-Mat: An Experimental Mutual Capacitance Dataset for Material Sensing and Inverse Problem Benchmarking

We introduce ECT-Mat, the first large-scale, experimentally collected dataset for planar Electrical Capacitance Tomography (ECT) under both contact and non-contact conditions. The dataset contains 18,432 high-resolution capacitance matrices acquired with a custom-designed 3×3 electrode sensor array, covering three representative materials (glass, resin, wood) and three geometric shapes (circle, triangle, square). Each sample includes raw mutual capacitance data, ground-truth object labels, and baseline Linear Back Projection (LBP) reconstructions. Unlike existing simulation-based benchmarks, ECT-Mat captures real-world measurement characteristics, including dielectric variability, positional uncertainty, and environmental noise, thereby bridging the gap between numerical modeling and practical sensing. To ensure reproducibility and reuse, all sensor fabrication files, acquisition codes, and calibration protocols are openly released. ECT-Mat provides a standardized benchmark for inverse problem research, supports machine learning–based image reconstruction, and enables applications in touchless material sensing and human–computer interaction. By combining scale, diversity, and openness, ECT-Mat establishes a foundation for advancing both physics-informed and data-driven approaches to capacitance tomography.