T-FAKE: Synthesizing Thermal Images for Facial Landmarking

Downloading the dataset

To download the color images, sparse annotations, and segmentation masks for the dataset, please use the links in the FaceSynthetics repository.

Our dataset has been generated for a warm and for a cold condition. Each dataset can be downloaded separately as

• A small sample with 100 images from here warm and here cold
• A medium sample with 1,000 images from here warm and here cold
• The full dataset with 100,000 images from here warm and here cold
• The dense annotations are available from here

Using the landmarker

A landmarker trained on the T-FAKE dataset is available via pip and from this repository: thermal-face-alignment.

Install and run:

pip install thermal-face-alignment

import cv2
from tfan import ThermalLandmarks

# Read a thermal image, normalized grayscale or temperature values:
image = cv2.imread("thermal.png", cv2.IMREAD_GRAYSCALE)

# Initialize landmarker (downloads weights on first use)
landmarker = ThermalLandmarks(device="cpu", n_landmarks=478)

landmarks, confidences = landmarker.process(image)

Predicted 70 and 478 point landmarks on an example from the BU-TIV Benchmark.

Pre-trained models

The models for the thermalization can be downloaded from here.

Thermalization

Our baseline U-Net translation model is imported from segmentation_models_pytorch library. Specifically, we define the translator as follows:

import segmentation_models_pytorch as smp

translator = smp.Unet(
    encoder_name="resnet34",        
    encoder_weights="imagenet",     
    in_channels=3,                  
    classes=1,                      
    activation="sigmoid"
)

This model is based on a U-Net architecture with a ResNet-34 encoder pre-trained on ImageNet. It takes three-channel RGB input images and outputs a single-channel thermal image with a sigmoid activation function. For training progress of the thermalization model see ThermalizationCode/ThermalizerOutput.ipynb.

Running the benchmark

To run the benchmark, you have to download the CHARLOTTE ThermalFace dataset.

License

This dataset and the landmarking methods are licensed under the Attribution-NonCommercial-ShareAlike 4.0 International license as it is derived from the FaceSynthetics dataset.

Citation

If you use this code for your own work, please cite our paper:

P. Flotho, M. Piening, A. Kukleva and G. Steidl, “T-FAKE: Synthesizing Thermal Images for Facial Landmarking,” Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR), 2025. CVF Open Access

BibTeX entry

@InProceedings{tfake2025_CVPR,
    author    = {Flotho, Philipp and Piening, Moritz and Kukleva, Anna and Steidl, Gabriele},
    title     = {T-FAKE: Synthesizing Thermal Images for Facial Landmarking},
    booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {26356-26366}
}