SF-TL54: A Thermal Facial Landmark Dataset with Visual Pairs

In this work, we present a thermal face dataset with manually annotated bounding boxes and facial landmarks. The dataset was constructed using our large-scale SpeakingFaces dataset ( In total, the dataset contains 2,556 thermal-visual image pairs of 142 subjects, where each subject has 18 thermal-visual image pairs (2 trial x 9 positions). The dataset, annotations, source code, and pre-trained thermal face detection and facial landmark prediction models are publicly available to advance research in thermal face analysis.

Examples of annotated thermal images for one subject.

The landmarks have the following order:

There are many potential applications, including, but not limited to:

Educational video tutorials:

GitHub icon
Powered by GitHub