A single linear layer which does a 1x1 conv

A single linear layer which does a 1x1 conv

Spatial transformer network. For optical flow based frame warping. Args: mode: sampling interpolation mode of `grid_sample` pad

Optical flow estimation network

Frame-recurrent network proposed in https://arxiv.org/abs/1801.04590

Metric calculator for model evaluation Currently supported metrics: * PSNR (RGB and Y) * LPIPS * tOF as

Residual block without batch normalization

Reconstruction & Upsampling network

A bicubic upsampling class with similar behavior to that in TecoGAN-Tensorflow Note that it's different from torch.nn.functional.interpolate

Charbonnier Loss (robust L1)

Originally from BasicSR: https://github.com/xinntao/BasicSR/blob/master/codes/models/lr_scheduler.py

takes 2 distances, puts through fc layers, spits out value between [0,1] (if use_sigmoid is True)

takes 2 distances, puts through fc layers, spits out value between [0,1] (if use_sigmoid is True)

Optical flow estimation network

Optical flow estimation network

Residual block without batch normalization

Residual block without batch normalization

Reconstruction & Upsampling network

Reconstruction & Upsampling network

Spatial discriminator

Spatio-Temporal discriminator in proposed in TecoGAN

A model wraper for subjective video super-resolution It contains a generator and a discriminator, as well as relative functions to t

A model wraper for objective video super-resolution It contains a generator, as well as relative functions to train and test the gen

Frame-recurrent network proposed in https://arxiv.org/abs/1801.04590

Frame-recurrent network proposed in https://arxiv.org/abs/1801.04590

Pre-trained network with all channels equally weighted by default

Spatial discriminator

Spatial discriminator

Spatio-Temporal discriminator in proposed in TecoGAN

Spatio-Temporal discriminator in proposed in TecoGAN