Published May 21, 2024 | Version v1
Dataset Open

UTKPAD: Replay Attack Database for Face Age Verification

  • 1. ROR icon Idiap Research Institute



UTKPAD is a replay attack database prepared for face age verification as part of the paper "Vulnerability of Face Age Verification to Replay Attacks" published in ICASSP 2024 conference. This database is originated from the well-known large dataset, UTKFace of face images with age labels ranging from 1 to 100 plus years old. Replay attack are recorded with three different mobiles phones: Apple iPhone 12, Samsung Galaxy S9 and Huawei Mate 30. With UTKPAD database, we also provide file lists that can be used for training and testing of replay attacks detection and for vulnerability assessment of age verification systems.

The face images in the UTKFace are first enhanced with a face restoration CodeFormer method. Each enhanced image is converted into a video clip with a subsequent concatenation process to have one video clip including all image-to-video converted files. The final video clip is then replayed on Apple iPad Pro in order to record it with three different mobiles phones: Apple iPhone 12, Samsung Galaxy S9 and Huawei Mate 30. And, finally, the recordings are de-concatenated/segmented into sub video clips and each video clip is sampled by taking the middle frame to construct the database of replay attack images.

To avoid breaching the copyright, we release the dataset in the form of deltas that are not actual images. The attack images can be recovered only if the user also obtains and downloads the original UTKFace dataset. When UTKFace is downloaded, however, the attack images of UTKPAD can be easily computed using the script we provide. 



If you're using this dataset, please cite the following publication

        author = {Korshunov, Pavel and George, Anjith and {\"O}zbulak, G{\"o}khan and Marcel, S{\'{e}}bastien},
        projects = {Idiap, Biometrics Center},
        title = {Vulnerablity of Face Age Verification to Replay Attacks},
        booktitle = {ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
        year = {2024},


Files (259.8 GB)

Name Size Download all
308.0 kB Download
39.1 GB Download
38.3 GB Download
11.4 GB Download
36.3 GB Download
35.1 GB Download
11.0 GB Download
38.9 GB Download
38.1 GB Download
11.8 GB Download

Additional details

Related works

Is described by
Conference proceeding: 10.1109/ICASSP48485.2024.10447255 (DOI)
Dataset: (Other)