File size: 957 Bytes
3ce19a2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 | import torch
import numpy as np
import imageio
def generate_rnd(H, sampler, shape, ema_imle, fname, logprint):
mb = H.num_rows_visualize
batches = []
n_rows = mb
temp_latent_rnds = torch.randn([mb, H.latent_dim], dtype=torch.float32).cuda()
for t in range(H.num_rows_visualize):
temp_latent_rnds.normal_()
if(H.use_snoise == True):
tmp_snoise = [s[:mb].normal_() for s in sampler.snoise_tmp]
else:
tmp_snoise = [s[:mb] for s in sampler.neutral_snoise]
# tmp_snoise = [s[:mb].normal_() for s in sampler.snoise_tmp]
out = ema_imle(temp_latent_rnds, tmp_snoise)
batches.append(sampler.sample_from_out(out))
im = np.concatenate(batches, axis=0).reshape((n_rows, mb, *shape[1:])).transpose([0, 2, 1, 3, 4]).reshape(
[n_rows * shape[1], mb * shape[2], 3])
logprint(f'printing samples to {fname}')
imageio.imwrite(fname, im)
|