from deliravision.models.gans import WassersteinDivergenceGAN from deliravision.losses import WassersteinDivergence 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 = 1000 key_mapping = {"x": "data"} torch.autograd.set_detect_anomaly(True) model, weight_path = train(WassersteinDivergenceGAN, {"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={"divergence": WassersteinDivergence()}) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
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)
from deliravision.models.gans import DRAGAN from deliravision.losses import GradientPenalty 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( DRAGAN, { "latent_dim": 100, "num_channels": 1, "img_size": 28 }, os.path.join(outpath, "train"), img_path, num_epochs=num_epochs, key_mapping=key_mapping, additional_losses={"gradient_penalty": GradientPenalty()}) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
from deliravision.models.gans import GenerativeAdversarialNetworks 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 = {"imgs": "data"} model, weight_path = train(GenerativeAdversarialNetworks, { "latent_dim": 100, "img_shape": (1, 28, 28) }, os.path.join(outpath, "train"), img_path, num_epochs=num_epochs, key_mapping=key_mapping) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
from deliravision.models.gans import WassersteinGradientPenaltyGAN from deliravision.losses import GradientPenalty 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 = 1000 key_mapping = {"x": "data"} torch.autograd.set_detect_anomaly(True) model, weight_path = train(WassersteinGradientPenaltyGAN, {"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={"gradient_penalty": GradientPenalty()}) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
from deliravision.models.gans import WassersteinGAN 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 = 1000 key_mapping = {"x": "data"} torch.autograd.set_detect_anomaly(True) model, weight_path = train(WassersteinGAN, { "latent_dim": 100, "img_shape": (1, 28, 28) }, os.path.join(outpath, "train"), img_path, num_epochs=num_epochs, key_mapping=key_mapping) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
from deliravision.models.gans import RelativisticGAN from deliravision.losses import AdversarialLoss 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 = 1000 key_mapping = {"x": "data"} model, weight_path = train(RelativisticGAN, { "latent_dim": 100, "img_size": 28, "num_channels": 1 }, os.path.join(outpath, "train"), img_path, num_epochs=num_epochs, key_mapping=key_mapping, additional_losses={ "adversarial": AdversarialLoss( torch.nn.BCEWithLogitsLoss()) }) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
batchsize = 64 latent_dim = 100 n_classes = 10 code_dim = 2 model, weight_path = train(InfoGAN, { "latent_dim": latent_dim, "num_channels": 1, "img_size": 28, "n_classes": n_classes, "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,
from deliravision.models.gans import DeepConvolutionalGAN 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 = {"imgs": "data"} model, weight_path = train(DeepConvolutionalGAN, {"latent_dim": 100, "img_size": 28, "num_channels": 1}, os.path.join(outpath, "train"), img_path, num_epochs=num_epochs, key_mapping=key_mapping) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
from deliravision.models.gans import BoundaryEquilibriumGAN from deliravision.losses import BELoss 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(BoundaryEquilibriumGAN, { "latent_dim": 100, "img_size": 28, "n_channels": 1 }, os.path.join(outpath, "train"), img_path, num_epochs=num_epochs, key_mapping=key_mapping, additional_losses={"began": BELoss()}) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
img_path = os.path.abspath("~/data/") 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
from deliravision.models.gans import ConditionalGAN from deliravision.losses import AdversarialLoss import torch 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", "labels": "label"} batchsize = 64 latent_dim = 100 n_classes = 10 model, weight_path = train(ConditionalGAN, {"latent_dim": latent_dim, "n_classes": n_classes, "img_shape": (1, 28, 28)}, os.path.join(outpath, "train"), img_path, num_epochs=num_epochs, key_mapping=key_mapping, additional_losses={"adversarial": AdversarialLoss( torch.nn.MSELoss())}, 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, 1))],)
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 = 1000 key_mapping = {"imgs": "data"} model, weight_path = train(EnergyBasedGAN, { "latent_dim": 100, "num_channels": 1, "img_size": 28 }, os.path.join(outpath, "train"), img_path, num_epochs=num_epochs, key_mapping=key_mapping, additional_losses={ "pullaway": PullAwayLoss(), "pixelwise": torch.nn.MSELoss(), "discriminator_margin": DiscriminatorMarginLoss() }) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
from deliravision.models.gans import LeastSquareGAN from deliravision.losses import AdversarialLoss 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 = 1000 key_mapping = {"imgs": "data"} model, weight_path = train( LeastSquareGAN, { "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={"adversarial": AdversarialLoss(torch.nn.MSELoss())}) predict(model, weight_path, os.path.join(outpath, "preds"), num_epochs)
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")