Parameters: :param lr_curr: the current lr data in shape nchw :param lr_prev: the previous lr data in sha

:param h: mask height :param w: mask width :param split_rate: 0.0 - 1.0 :return: mask zero_one matirx bottom-left

INPUTS model - ['net-lin'] for linearly calibrated network ['net'] for off-the-shelf network

pad tensor if its width and height couldn't be divided by `value`. Args: x: a tensor at least has 3 dimensions. value: value to d

prepare gt, lr data for training for BD degradation, generate lr data and remove border of gt data for BI degradation, return data d

Function of mini-batch training Parameters: :param data: a batch of training data (lr & gt) in shape ntchw

Coarse to fine flow estimation network Originally from paper "Real-Time Video Super-Resolution with Spatio-Temporal Networks and Moti

Flow estimation network Originally from paper "FlowNet: Learning Optical Flow with Convolutional Networks" and adapted according to p

Folder dataset with paired data support both BI & BD degradation

LMDB dataset with paired data, for BI degradation

LMDB dataset with unpaired data, for BD degradation