Separate training

[JohannaSpr]

In your paper, you mention that you experimented also with separated training. In training_options.py is an argument --separate, but I could not find it anywhere in the code! Could you explain how the separated training works?

[lyndonzheng]

@JohannaSpr This is our missing. Since the separate training always get worse result, that's we delete this separate training function in the final multi-gpu version. If you want to try separate training, an easy way is to set the "---lambda_rec_lab","--lambda_gan_feature" and "--lambda_smooth" to 0. Then, using this pre-trained translation network to train the task network.

[JohannaSpr]

Thanks for the fast answer!