Simulated Three-Frequency Phase Unwrapping Dataset with Ground-Truth Absolute Phase

This dataset is a simulated three-frequency phase unwrapping dataset generated for phase unwrapping research and algorithm evaluation. Due to the difficulty of obtaining accurate ground-truth absolute phase data in real optical measurement systems, the data in this dataset are generated synthetically to provide reliable and controllable ground truth for both traditional and deep learning–based methods.The absolute phase maps are generated using two synthetic phase generation methods. One method is based on an improved random matrix enlargement technique, in which small random matrices are first created and then upsampled to a fixed resolution of 256×256 using bilinear interpolation to form smooth phase distributions. In addition, Gaussian function–based generation is also adopted to produce phase maps with specific characteristics, such as local phase variations and discontinuities. These two methods enable the generation of diverse phase patterns with different spatial properties. The absolute phase values are normalized to a predefined range and optionally corrupted with Gaussian noise of different strengths to simulate realistic measurement conditions.Based on each absolute phase map, three wrapped phase maps with linearly spaced frequencies are generated to simulate a traditional three-frequency phase measurement system. Noise with predefined signal-to-noise ratios is added to the wrapped phase maps to obtain both noise-free and noisy samples. The dataset contains 16,500 samples. For each sample, absolute phase maps (ground truth) and wrapped phase maps at three frequencies are provided.This dataset can be used for training, validation, and quantitative evaluation of phase unwrapping algorithms, as well as for comparative studies between traditional multi-frequency methods and learning-based approaches.