TFD68: A Fully Annotated Thermal Facial Dataset with 68 Landmarks, Pose Variations, Per-Pixel Thermal Maps, Visual Pairs, Occlusions, and Facial Expressions

Facial analysis in the thermal infrared spectrum enables low-light operation, medical diagnostics, and privacy-preserving biometrics, but scarce annotated datasets limit progress. We introduce TFD68: 28,496 thermal images of 137 individuals, each paired with a visual image matched for head poses, occlusions, and expression. Every thermal image includes 68 facial landmarks and per-pixel thermal maps. The dataset contains systematic yaw and pitch variations, accessory occlusions, and a wide range of expressions to support pose-invariant, occlusion-resilient evaluation. We demonstrate TFD68’s utility on thermal facial landmark detection and facial expression recognition, and show that it enables cross-modal and physiological analyses. To our knowledge, TFD68 is the first thermal facial dataset combining detailed landmark annotations, visual-thermal pairs, and temperature maps, establishing a foundation for advanced thermal facial research.