dl = dl.reshape((latent_size, -1)).transpose()
else:
dl = dl.reshape((-1, latent_size))
else:
ql = latents
dl = latents
if args.transpose:
dl = dl.reshape((latent_size, -1)).transpose()
else:
dl = dl.reshape((-1, latent_size))
with torch.no_grad():
if args.gpu:
if args.split == 0:
predict = test.decode(
torch.from_numpy(dl).to(device)).cpu().detach().numpy()
else:
split = args.split
len_dl = dl.shape[0]
start = 0
predict = None
while (start < len_dl):
end = min(len_dl, start + split)
dl_split = dl[start:end]
predict_s = test.decode(
torch.from_numpy(dl_split).to(
device)).cpu().detach().numpy()
if start == 0:
predict = predict_s
else:
predict = np.concatenate((predict, predict_s))
if args.mode != "d":
with torch.no_grad():
#try:
outputs = test(torch.from_numpy(picts).to(device))
zs = outputs[2].cpu().detach().numpy()
predict = outputs[0].cpu().detach().numpy()
else:
zs = np.fromfile(args.latents, dtype=np.float32)
if args.transpose:
zs = zs.reshape((args.lsize, -1)).transpose()
else:
zs = zs.reshape((-1, args.lsize))
with torch.no_grad():
if args.gpu:
predict = test.decode(
torch.from_numpy(zs).to(device)).cpu().detach().numpy()
else:
predict = test.decode(torch.from_numpy(zs)).detach().numpy()
#predict=outputs[0].numpy()
print(zs.shape)
#print(predict.size)
qs = []
us = []
recon = np.zeros((height, width), dtype=np.float32)
eb = args.error * rng
if args.normalize:
picts = (picts + 1) / 2
picts = picts * (global_max - global_min) + global_min
predict = (predict + 1) / 2
predict = predict * (global_max - global_min) + global_min