Recursively finds all files matching the pattern.

output: [batch_size, h, w, c] images, scaled to [0., 1.] ***[BUG] shuffle=False doesn't seem to work correctly!!! *****

Save a model to logdir/model.ckpt-[step]

Kullback-Leibler divergence of two Gaussians *Assuming that each dimension is independent mu: mean lv: log variance

[TODO] Although most of the trainer structures are the same, I think we have to use different training scripts for VAE- and DC-GAN

Interpolation from the latent space

Try to load model form a dir (search for the newest checkpoint)

Note: 1. The input is rescaled to [-1, 1] (img_reader: rtype)

Generate fake samples given `z` if z is not given or is an `int`, this fcn generates (z=128) samples

Leaky ReLU

truth: 0. or 1.