About generating image - Githubissues

grey-eye / talking-heads

Our implementation of "Few-Shot Adversarial Learning of Realistic Neural Talking Head Models" (Egor Zakharov et al.)

GNU General Public License v3.0

593 stars 109 forks source link

About generating image #29

Closed LWei-gfkd closed 5 years ago

LWei-gfkd commented 5 years ago

The generated image looks like a 4*4 image, and you can't see the original avatar at all.And after iterations hundreds of times, the loss value becomes nan.

castelo-software commented 5 years ago

@LWei-gfkd. As the readme explains, I haven't gotten around to implementing fine-tuning. For now only pre-training with the feed forward model has been finished. If you want to help us implement it, feel free to update us on your findings here https://github.com/grey-eye/talking-heads/issues/26

If you just want to play with the feed forward model, change the FEED_FORWARD = False in the config file to True.