search

LoSealL / VideoSuperResolution

A collection of state-of-the-art video or single-image super-resolution architectures, reimplemented in tensorflow.
MIT License
1.61k stars 295 forks source link

Training of TecoGAN hasn't been verified #95

Closed Feihong-cc closed 4 years ago

Feihong-cc commented 4 years ago

Hi, I tried to train and test the TecoGAN model, but it doesn't seem to be supported. Have you completed this part?

LoSealL commented 4 years ago

not yet

Feihong-cc commented 4 years ago

oh, thank you very much! what has and has not yet been implemented now?

LoSealL commented 4 years ago

The TecoGAN networks and training are implemented followed by the original paper. But the hyper-parameters are not tuned so the training will not converge.

Now the author open-sourced his codes including training script. You can refer his code