shaoanlu

Auto image scaling is used to prevent OOM error for face detection and alignment.

The iris detector I used in this project seems not performing well due to [its deviated pre-processing from official implementation](https://github.com/shaoanlu/GazeML-keras#wip). Disabling `draw_iris()` function in [`utils.py`](https://github.com/shaoanlu/fewshot-face-translation-GAN/blob/master/utils/utils.py) might help. Also, the face...

The unnatural contrast and highlight in the output faces are the intrinsic characteristics of the released model. These artifacts can also be observed in the readme figures. I did not...

I might not have to time to update the code until mid or late Oct. 😔.

The author of FUNIT actually showed results on face identity translation in [this video](https://youtu.be/IhAsXcCz8LI?t=1220). I've also trained a FUNIT model on VGGFace2 dataset using the official implementation. The figure below...

To discriminate if the two input images belong to the same identity, i.e., for discriminating real/fake pairs.

There are actually real A pairs involved in training but are not shown in the README figure.

An [early work](https://arxiv.org/abs/1803.11182) by Bao et al. has vary similar approach with ours. Content/style disentanglement fashion has been used in many papers for image translation, e.g.[ DRIT](https://github.com/HsinYingLee/DRIT) and [MUNIT](https://github.com/NVlabs/MUNIT), with...

Downgrading tensorflow and keras might solve this problem. I'm not sure whether downgrading will cause other problems since the environment in colab has changed a lot since last time I...

I just found this: https://github.com/plaidml/plaidml https://twitter.com/fchollet/status/921815941112930304?lang=en It uses OpenCL and adapts keras code.