RealDigiFace
Description
The more realistic version of the DigiFace1M.
DigiFace1M originally created by the 3D rendering engines. When this dataset is used to train Face Recognition systems there is OOD problem by evaluating the trained models against real dataset like LFW, IJB-B and IJB-C. In our paper we tried to address this problem by post processing the original images.
This type of dataset is generated by simply passing the images from DigiFace1M through methods like CodeFormer and VSAIT.
License
Digi2Real dataset is published for non-commercial research use only, please check the license agreement.
Reference
If you use this dataset, please cite the following publication:
Rahimi, P., Razeghi, B., & Marcel, S. (2024). Synthetic to Authentic: Transferring Realism to 3D Face Renderings for Boosting Face Recognition. ArXiv, abs/2407.07627. European Conference on Computer Vision Workshop of Synthetic Data in Computer Vision, September 2024