A new arXiv preprint introduces RoboCloth, a dataset of 500 real cloth materials captured for neural reflectance reconstruction. The acquisition process uses robotically controlled cameras and lighting, allowing flexible yet precisely calibrated configurations. Each material is imaged in a setup that can be adjusted without losing calibration, which is a key advantage for capturing diverse fabric appearances.
The dataset is positioned as a large-scale resource for training and evaluating neural reflectance models. By providing real cloth samples rather than synthetic approximations, it addresses a common gap in material capture research. The robotic control also enables consistent, repeatable imaging across different lighting and viewing conditions.
The abstract does not detail the exact imaging time per material or the range of cloth types included, but the scale and robotic acquisition are the main contributions. RoboCloth appears intended as a practical benchmark for improving how digital systems reconstruct the look of real fabrics.