from deliravision.models.gans import BoundarySeekingGAN
from deliravision.losses import BoundarySeekingLoss
import os
from training.gans._basic import train, predict
if __name__ == '__main__':
img_path = os.path.abspath("~/data/")
outpath = os.path.abspath("~/GanExperiments")
num_epochs = 1000
key_mapping = {"x": "data"}
model, weight_path = train(BoundarySeekingGAN, {"latent_dim": 100,
"img_shape": (1, 28, 28)},
os.path.join(outpath, "train"), img_path,
num_epochs=num_epochs,
key_mapping=key_mapping,
additional_losses={"boundary_seeking":
BoundarySeekingLoss()})
predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
outpath = os.path.abspath("~/GanExperiments")
num_epochs = 1000
key_mapping = {"real_imgs": "data", "real_labels": "label"}
latent_dim = 100
batchsize = 64
n_classes = 10
model, weight_path = train(
AuxiliaryClassifierGANPyTorch, {
"latent_dim": latent_dim,
"img_size": 28,
"n_channels": 1,
"n_classes": n_classes
},
os.path.join(outpath, "train"),
img_path,
num_epochs=num_epochs,
additional_losses={"auxiliary": torch.nn.CrossEntropyLoss()},
key_mapping=key_mapping,
batchsize=batchsize)
predict(model,
weight_path,
os.path.join(outpath, "preds"),
num_epochs,
gen_fns=[torch.randn, torch.randint],
gen_args=[(batchsize, latent_dim), (0, n_classes, (batchsize, ))],
gen_kwargs=[{}, {
"dtype": torch.long
}])
"code_dim": code_dim
},
os.path.join(outpath, "train"),
img_path,
num_epochs=num_epochs,
key_mapping=key_mapping,
additional_losses={
"categorical":
torch.nn.CrossEntropyLoss(),
"continuous":
torch.nn.MSELoss(),
"adversarial":
AdversarialLoss(torch.nn.MSELoss())
},
create_optim_fn=create_optims,
batchsize=batchsize)
predict(model,
weight_path,
os.path.join(outpath, "preds"),
num_epochs,
gen_fns=[torch.randn, onehot_ints, uniform],
gen_args=[(batchsize, latent_dim), (n_classes, (batchsize, 1)),
((batchsize, code_dim), )],
gen_kwargs=[{}, {
"dtype": torch.long
}, {
"min_val": -1,
"max_val": 1
}])
from deliravision.models.gans import AdversarialAutoEncoderPyTorch
import os
from training.gans._basic import train, predict
import torch
if __name__ == '__main__':
img_path = os.path.abspath("~/data/")
outpath = os.path.abspath("~/GanExperiments")
num_epochs = 1500
key_mapping = {"x": "data"}
model, weight_path = train(
AdversarialAutoEncoderPyTorch, {
"latent_dim": 100,
"img_shape": (1, 28, 28)
},
os.path.join(outpath, "train"),
img_path,
num_epochs=num_epochs,
additional_losses={"pixelwise": torch.nn.L1Loss()},
key_mapping=key_mapping)
predict(model,
weight_path,
os.path.join(outpath, "preds"),
num_epochs,
generative_network="generator.decoder")