Two-layer 2D convolutional block with a PReLU activation in between.

Residual Dense Block. Used in RRDB block in ESRGAN. Args: num_feat (int): Channel number of intermediate features. num_grow_

A block that fuses two revisits into one.

Two-layer 2D convolutional block (DoubleConv2d) with a skip-connection to a sum.

WeightedRandomSampler except allows for more than 2^24 samples to be sampled. Source code: https://github.com/pytorch/pytorch/issues/2576#iss

High-resolution CNN. Inherits as many elements from SRCNN as possible for as fair a comparison: - DoubleConv2d: the in_channels are doubled b

One-hot encoder.

PixelShuffle block with ConvTranspose2d for sub-pixel convolutions.

Recursively fuses a set of encodings.

Super-resolution CNN. Uses no recursive function, revisits are treated as channels.

Networks consisting of Residual in Residual Dense Block, which is used in ESRGAN. ESRGAN: Enhanced Super-Resolution Generative Adversarial Ne

Upsampler block with ConvTranspose2d.

Modulated deformable conv for deformable alignment. Different from the official DCNv2Pack, which generates offsets and masks from the precedi

Two-layer 2D convolutional block (DoubleConv2d) with a skip-connection to a concatenation (instead of a sum used in ResidualBlock).

Wrapper model code to run the SRCNN and HighResNet architectures. Loss weights taken from the WorldStrat paper. Losses are hardcoded.

Arg: num_in_ch (int): Channel number of inputs. Default: 3. num_feat (int): Channel number of base intermediate features. Default

SSR ESRGAN Model: Training Satellite Imagery Super Resolution with Paired Training Data. The input to the generator is a time series of Sent

Residual in Residual Dense Block. Used in RRDB-Net in ESRGAN. Args: num_feat (int): Channel number of intermediate features.

Residual block without BN. Args: num_feat (int): Channel number of intermediate features. Default: 64. res_scale (flo

Dataset object for the S2NAIP data. Builds a list of Sentinel-2 time series and NAIP image pairs. Args: opt (dict): Config for train

SSR ESRGAN Model: Satellite imagery super-resolution model based on the Real-ESRGAN architecture. The input to the generator is a time serie

Defines a U-Net discriminator with spectral normalization (SN) It is used in Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Sy

Upsample module. Args: scale (int): Scale factor. Supported scales: 2^n and 3. num_feat (int): Channel number of intermediate fea