IJCB 2021 - Generation of synthetic face recognition datasets using StyleGAN2
=============
=============
This package is part of the signal-processing and machine learning toolbox Bob_. It contains source code to reproduce experiments published in the following article::
This package is part of the signal-processing and machine learning toolbox Bob_. It contains source code to reproduce experiments published in the following article::
...
@@ -49,7 +49,7 @@ License and usage
...
@@ -49,7 +49,7 @@ License and usage
This package is released under the `BSD-3 license <https://gitlab.idiap.ch/bob/bob.paper.ijcb2021_synthetic_dataset/-/blob/master/LICENSE>`_. It makes uses of some components from the
This package is released under the `BSD-3 license <https://gitlab.idiap.ch/bob/bob.paper.ijcb2021_synthetic_dataset/-/blob/master/LICENSE>`_. It makes uses of some components from the
`official release of the StyleGAN2 model <https://github.com/NVlabs/stylegan2>`_, which is itself released under the `Nvidia Source Code License-NC <https://gitlab.idiap.ch/bob/bob.paper.ijcb2021_synthetic_dataset/-/blob/master/bob/paper/ijcb2021_synthetic_dataset/stylegan2/LICENSE.txt>`_.
`official release of the StyleGAN2 model <https://github.com/NVlabs/stylegan2>`_, which is itself released under the `Nvidia Source Code License-NC <https://gitlab.idiap.ch/bob/bob.paper.ijcb2021_synthetic_dataset/-/blob/master/bob/paper/ijcb2021_synthetic_dataset/stylegan2/LICENSE.txt>`_.
Make sure to apply limitations of *both* licenses when reusing this work.
Make sure to consider limitations induced by *both* licenses when reusing this work.
Mainly, the StyleGAN2 code base can only be used for *research and evaluation purpose only*, therefore this project
Mainly, the StyleGAN2 code base can only be used for *research and evaluation purpose only*, therefore this project
itself cannot be used for anything else.
itself cannot be used for anything else.
...
@@ -91,7 +91,7 @@ Download model dependencies
...
@@ -91,7 +91,7 @@ Download model dependencies
The database generation in this project relies on several preexisting pretrained models:
The database generation in this project relies on several preexisting pretrained models:
* **DLIB Face Landmark detector** for cropping and aligning the projected faces exactly as in FFHQ. (`Example <http://dlib.net/face_landmark_detection.py.html>`_)
* **DLIB Face Landmark detector** for cropping and aligning the projected faces exactly as in FFHQ. (`Example <http://dlib.net/face_landmark_detection.py.html>`_)
* **StyleGAN2** as the main face synthesis network. ([Original paper](https://arxiv.org/abs/1912.04958), `Official repository <https://github.com/NVlabs/stylegan2>`_. We are using Config-F, trained on FFHQ at resolution 1024 x 1024
* **StyleGAN2** as the main face synthesis network. (`Original paper <https://arxiv.org/abs/1912.04958>`_, `Official repository <https://github.com/NVlabs/stylegan2>`_). We are using Config-F, trained on FFHQ at resolution 1024 x 1024
* A pretrained **VGG16** model, used to compute a perceptual loss between projected and target image (`Original paper <https://arxiv.org/abs/1801.03924>`_)
* A pretrained **VGG16** model, used to compute a perceptual loss between projected and target image (`Original paper <https://arxiv.org/abs/1801.03924>`_)
* A pretrained face recognition network (Inception-Resnet v2 trained on MSCeleb), to compute the embedding distance between identities in order to apply the ICT constraint.
* A pretrained face recognition network (Inception-Resnet v2 trained on MSCeleb), to compute the embedding distance between identities in order to apply the ICT constraint.
