mmagic.models.editors.lsgan

Package Contents

Classes

LSGAN	Implementation of Least Squares Generative Adversarial Networks.
LSGANDiscriminator	Discriminator for LSGAN.
LSGANGenerator	Generator for LSGAN.

class mmagic.models.editors.lsgan.LSGAN(generator: ModelType, discriminator: Optional[ModelType] = None, data_preprocessor: Optional[Union[dict, mmengine.Config]] = None, generator_steps: int = 1, discriminator_steps: int = 1, noise_size: Optional[int] = None, ema_config: Optional[Dict] = None, loss_config: Optional[Dict] = None)

Bases: mmagic.models.base_models.BaseGAN

Implementation of Least Squares Generative Adversarial Networks.

Paper link: https://arxiv.org/pdf/1611.04076.pdf

Detailed architecture can be found in LSGANGenerator and LSGANDiscriminator

disc_loss(disc_pred_fake: torch.Tensor, disc_pred_real: torch.Tensor) → Tuple

Get disc loss. LSGAN use the least squares loss to train the discriminator.

\[L_{D}=\left(D\left(X_{\text {data }}\right)-1\right)^{2} +(D(G(z)))^{2}\]

Parameters

• disc_pred_fake (Tensor) – Discriminator’s prediction of the fake images.

• disc_pred_real (Tensor) – Discriminator’s prediction of the real images.

Returns

Loss value and a dict of log variables.

Return type

tuple[Tensor, dict]

gen_loss(disc_pred_fake: torch.Tensor) → Tuple

Get gen loss. LSGAN use the least squares loss to train the generator.

\[L_{G}=(D(G(z))-1)^{2}\]

Parameters

disc_pred_fake (Tensor) – Discriminator’s prediction of the fake images.

Returns

Loss value and a dict of log variables.

Return type

tuple[Tensor, dict]

train_discriminator(inputs: dict, data_samples: mmagic.structures.DataSample, optimizer_wrapper: mmengine.optim.OptimWrapper) → Dict[str, torch.Tensor]

Train discriminator.

Parameters

• inputs (dict) – Inputs from dataloader.

• data_samples (DataSample) – Data samples from dataloader.

• optim_wrapper (OptimWrapper) – OptimWrapper instance used to update model parameters.

Returns

A dict of tensor for logging.

Return type

Dict[str, Tensor]

train_generator(inputs: dict, data_samples: List[mmagic.structures.DataSample], optimizer_wrapper: mmengine.optim.OptimWrapper) → Dict[str, torch.Tensor]

Train generator.

Parameters

• inputs (dict) – Inputs from dataloader.

• data_samples (List[DataSample]) – Data samples from dataloader. Do not used in generator’s training.

• optim_wrapper (OptimWrapper) – OptimWrapper instance used to update model parameters.

Returns

A dict of tensor for logging.

Return type

Dict[str, Tensor]

class mmagic.models.editors.lsgan.LSGANDiscriminator(input_scale=128, output_scale=8, out_channels=1, in_channels=3, base_channels=64, conv_cfg=dict(type='Conv2d'), default_norm_cfg=dict(type='BN'), default_act_cfg=dict(type='LeakyReLU', negative_slope=0.2), out_act_cfg=None, init_cfg=None)

Bases: mmengine.model.BaseModule

Discriminator for LSGAN.

Implementation Details for LSGAN architecture:

1. Adopt convolution in the discriminator;

2. Use batchnorm in the discriminator except for the input and final output layer;

3. Use LeakyReLU in the discriminator in addition to the output layer;

4. Use fully connected layer in the output layer;

5. Use 5x5 conv rather than 4x4 conv in DCGAN.

Parameters

• input_scale (int, optional) – The scale of the input image. Defaults to 128.

• output_scale (int, optional) – The final scale of the convolutional feature. Defaults to 8.

• out_channels (int, optional) – The channel number of the final output layer. Defaults to 1.

• in_channels (int, optional) – The channel number of the input image. Defaults to 3.

• base_channels (int, optional) – The basic channel number of the generator. The other layers contains channels based on this number. Defaults to 128.

• conv_cfg (dict, optional) – Config for the convolution module used in this discriminator. Defaults to dict(type=’Conv2d’).

• default_norm_cfg (dict, optional) – Norm config for all of layers except for the final output layer. Defaults to dict(type='BN').

• default_act_cfg (dict, optional) – Activation config for all of layers except for the final output layer. Defaults to dict(type='LeakyReLU', negative_slope=0.2).

• out_act_cfg (dict, optional) – Activation config for the final output layer. Defaults to dict(type='Tanh').

• init_cfg (dict, optional) – Initialization config dict.

forward(x)

Forward function.

Parameters

x (torch.Tensor) – Fake or real image tensor.

Returns

Prediction for the reality of the input image.

Return type

torch.Tensor

class mmagic.models.editors.lsgan.LSGANGenerator(output_scale=128, out_channels=3, base_channels=256, input_scale=8, noise_size=1024, conv_cfg=dict(type='ConvTranspose2d'), default_norm_cfg=dict(type='BN'), default_act_cfg=dict(type='ReLU'), out_act_cfg=dict(type='Tanh'), init_cfg=None)

Bases: mmengine.model.BaseModule

Generator for LSGAN.

Implementation Details for LSGAN architecture:

1. Adopt transposed convolution in the generator;

2. Use batchnorm in the generator except for the final output layer;

3. Use ReLU in the generator in addition to the final output layer;

4. Keep channels of feature maps unchanged in the convolution backbone;

5. Use one more 3x3 conv every upsampling in the convolution backbone.

We follow the implementation details of the origin paper: Least Squares Generative Adversarial Networks https://arxiv.org/pdf/1611.04076.pdf

Parameters

• output_scale (int, optional) – Output scale for the generated image. Defaults to 128.

• out_channels (int, optional) – The channel number of the output feature. Defaults to 3.

• base_channels (int, optional) – The basic channel number of the generator. The other layers contains channels based on this number. Defaults to 256.

• input_scale (int, optional) – The scale of the input 2D feature map. Defaults to 8.

• noise_size (int, optional) – Size of the input noise vector. Defaults to 1024.

• conv_cfg (dict, optional) – Config for the convolution module used in this generator. Defaults to dict(type=’ConvTranspose2d’).

• default_norm_cfg (dict, optional) – Norm config for all of layers except for the final output layer. Defaults to dict(type=’BN’).

• default_act_cfg (dict, optional) – Activation config for all of layers except for the final output layer. Defaults to dict(type=’ReLU’).

• out_act_cfg (dict, optional) – Activation config for the final output layer. Defaults to dict(type=’Tanh’).

• init_cfg (dict, optional) – Initialization config dict.

forward(noise, num_batches=0, return_noise=False)

Forward function.

Parameters

• noise (torch.Tensor | callable | None) – You can directly give a batch of noise through a torch.Tensor or offer a callable function to sample a batch of noise data. Otherwise, the None indicates to use the default noise sampler.

• num_batches (int, optional) – The number of batch size. Defaults to 0.

• return_noise (bool, optional) – If True, noise_batch will be returned in a dict with fake_img. Defaults to False.

Returns

If not return_noise, only the output image

will be returned. Otherwise, a dict contains fake_img and noise_batch will be returned.

Return type

torch.Tensor | dict