DIRT-SR38: Real Paired Cross-Sensor Thermal Sequence Dataset

Dataset DIRT-SR38 is a real paired cross-sensor thermal sequence dataset for ×2 super-resolution in clinical dynamic infrared thermography (DIRT). The dataset contains 30,780 aligned low-resolution (LR) and high-resolution (HR) thermal image pairs from 38 subjects acquired under a standardized abdominal cooling–rewarming protocol. Data are provided as intensity-normalized 8-bit and 16-bit PNG images. Pixel values represent relative thermal intensity and do not correspond to absolute temperature measurements. The dataset is intended for research in thermal image super-resolution, cross-sensor modeling, and dynamic infrared thermography analysis. For a detailed description of the dataset, acquisition protocol, LR–HR alignment procedure, preprocessing pipeline, super-resolution task definition, and benchmark experiments, please refer to the accompanying paper: Clarys, W., Verspeek, S., De Kerf, T., Verstockt, J., Vanlanduit, S., Thiessen, F., & Steenackers, G. (2026). DIRT-SR38: A Real Paired Cross-Sensor Thermal Sequence Dataset for Super-Resolution in Clinical Dynamic Infrared Thermography. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 7048–7057. Paper PDF: https://openaccess.thecvf.com/content/CVPR2026W/PBVS/papers/Clarys_DIRT-SR38_A_Real_Paired_Cross-Sensor_Thermal_Sequence_Dataset_for_Super-Resolution_CVPRW_2026_paper.pdf Citation If you use DIRT-SR38 in your research, please cite both the dataset record and the accompanying paper. Dataset: Clarys, W., Verspeek, S., De Kerf, T., Verstockt, J., Vanlanduit, S., Thiessen, F., & Steenackers, G. (2026). DIRT-SR38: Real Paired Cross-Sensor Thermal Sequence Dataset (Version 0.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.19241293 Paper: Clarys, W., Verspeek, S., De Kerf, T., Verstockt, J., Vanlanduit, S., Thiessen, F., & Steenackers, G. (2026). DIRT-SR38: A Real Paired Cross-Sensor Thermal Sequence Dataset for Super-Resolution in Clinical Dynamic Infrared Thermography. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 7048–7057. The dataset DOI should be cited to identify the exact data version used, while the paper should be cited for the dataset description, acquisition protocol, alignment/preprocessing pipeline, super-resolution task definition, and benchmark context.