Tongue Image Dataset

Here, we present a publicly available dataset containing 1194 tongue images captured under controlled conditions; the corresponding annotations (i.e., tongue color, tongue shape, cracked tongue, tooth-marked tongue, color of tongue coating, and thickness of tongue coating) assigned by three specialist TCM practitioners; and inquiry data collected from a clinical questionnaire. To ensure the reliability of these annotations of tongue characteristics, an annotation consistency evaluation was conducted.

Additionally, we designed and utilized deep learning models with corresponding single-label and multi-head classifiers to perform baseline validation on this dataset, highlighting the challenges and potential for future research. To the best of our knowledge, this is the first multi-label tongue image dataset. The use of this dataset can facilitate multidimensional diagnosis of physique, thereby promoting standardization of TCM diagnostic methods.