ResNet Variants Parameters ---------- block : Block Class for the residual block. Options are BasicBlockV1, BottleneckV1. lay

Standard convolution block with Batch normalization and activation.

r"""Randomly zeroes 2D spatial blocks of the input tensor. As described in the paper `DropBlock: A regularization method for convolutional net

Computes and stores the average and current value

Squeeze-and-Excitation block from 'Squeeze-and-Excitation Networks,' https://arxiv.org/abs/1709.01507.

A named tuple describing a ResNet block.

Profile the CUDA memory usage info for each line in pytorch This class registers callbacks for added functions to profiling them line by line

Convolution block with Batch normalization and ReLU pre-activation.

Cosine lr decay function with warmup. Lr warmup is proposed by ` Accurate, Large Minibatch SGD:Training ImageNet in 1 Hour` `http

DenseNet model from 'Densely Connected Convolutional Networks,' https://arxiv.org/abs/1608.06993.

DenseNet unit. Parameters: ---------- in_channels : int Number of input channels. out_channels : int Number of o

EfficientNet specific depthwise separable convolution block/unit with BatchNorms and activations at each convolution layers. Parameters:

EfficientNet specific initial block. Parameters: ---------- in_channels : int Number of input channels. out_channels : i

EfficientNet inverted residual unit. Parameters: ---------- in_channels : int Number of input channels. out_channels : i

EfficientNet model from 'EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,' https://arxiv.org/abs/1905.11946. Pa

Approximated sigmoid function, so-called hard-version of sigmoid from 'Searching for MobileNetV3,' https://arxiv.org/abs/1905.02244.

H-Swish activation function from 'Searching for MobileNetV3,' https://arxiv.org/abs/1905.02244. Parameters: ---------- inplace : boo

Identity block.

PyTorch implementation of the lookahead wrapper. Lookahead Optimizer: https://arxiv.org/abs/1907.08610

A memory reporter that collects tensors and memory usages Parameters: - model: an extra nn.Module can be passed to infer the name

ProxylessNAS block for residual path in ProxylessNAS unit. Parameters: ---------- in_channels : int Number of input channels

ProxylessNAS model from 'ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware,' https://arxiv.org/abs/1812.00332.

ProxylessNAS unit. Parameters: ---------- in_channels : int Number of input channels. out_channels : int Number

Split-Attention Conv2d

Swish activation function from 'Searching for Activation Functions,' https://arxiv.org/abs/1710.05941.

DenseNet's auxiliary block, which can be treated as the initial part of the DenseNet unit, triggered only in the first unit of each stage.

r"""Implement of AirFace:Lightweight and Efficient Model for Face Recognition (https://arxiv.org/pdf/1907.12256.pdf): Args:

r"""Implement of Am_softmax (https://arxiv.org/pdf/1801.05599.pdf): Args: in_features: size of each input sample out_features: siz

r"""Implement of Towards Flops-constrained Face Recognition (https://arxiv.org/pdf/1909.00632.pdf): Args: in_features: size of each input

ResNet Bottleneck

r"""Implement of ArcFace (https://arxiv.org/pdf/1801.07698v1.pdf): Args: in_features: size of each input sample out_fe

r"""Implement of CosFace (https://arxiv.org/pdf/1801.09414.pdf): Args: in_features: size of each input sample out_features: size o

A class to yield CUDA memory at any time in the training The whole save/load is a bit tricky because all data transfer should be inplace oper

r"""Implement of CurricularFace (https://arxiv.org/pdf/2004.00288.pdf): Args: in_features: size of each input sample out_features: