Convolution + IBNorm + ReLu

Fully connection operations for target net Note: Weights & biases are different for different images in a batch, thus here we

CoordConv layer as in the paper.

Residual block recommended in: http://torch.ch/blog/2016/02/04/resnets.html

A named tuple describing a ResNet block.

Helper for the face restoration pipeline (base class).

Fusion Branch of MODNet

High Resolution Branch of MODNet

Combine the hypernet and target network within a network.

Hyper network for learning perceptual rules. Args: lda_out_channels: local distortion aware module output size. hyper_in_chan

Combine Instance Norm and Batch Norm into One Layer

This class represents the internal state of individual tracked objects observed as bbox. doc: https://filterpy.readthedocs.io/en/latest/kalman

Low Resolution Branch of MODNet

Architecture of MODNet

MobileNetV2 Backbone

Normalization Layers. Args: channels: input channels, for batch norm and instance norm. input_size: input shape without batch siz

Relu Layer. Args: relu type: type of relu layer, candidates are - ReLU - LeakyReLU: default relu slope 0.2

SE Block Proposed in https://arxiv.org/pdf/1709.01507.pdf

SORT: A Simple, Online and Realtime Tracker. Ref: https://github.com/abewley/sort

Target network for quality prediction.

Superclass of Replaceable Backbone Model for Semantic Estimation