Conv-Norm-Activation block. Args: in_channels (int): Same as nn.Conv2d. out_channels (int): Same as nn.Conv2d. kernel_siz

For torch < 1.4 A wrapper around :class:`torch.nn.BatchNorm2d` to support zero-size tensor.

Computes and stores the average and current value

Computes and stores the average and current value

Computes and stores the average and current value

conv => norm => activation use native nn.Conv2d, not slimmable

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Subclass torch's LayerNorm to handle fp16.

Model Exponential Moving Average Keep a moving average of everything in the model state_dict (parameters and buffers). This is intended to al

Computes and stores the average and current value

Cityscapes semantic segmentation dataset. Arguments: root: Str, root directory. split: Str, data split, e.g. train/val/test.

Mixed Grouped Convolution Based on MDConv and GroupedConv in MixNet impl: https://github.com/tensorflow/tpu/blob/master/models/official/mna

Mixup/Cutmix that applies different params to each element or whole batch Args: mixup_alpha (float): mixup alpha value, mixup is active

AvgPool or MaxPool

Selectable global pooling layer with dynamic input kernel size

change the short edge length to size

Normalize a tensor image with mean and standard deviation. Args: mean (tuple): means for each channel. std (tuple): standard devia

Convert ndarrays in sample to Tensors.

(Dilated) depthwise separable conv ReLU - (Dilated) depthwise separable - Pointwise - BN If dilation == 2, 3x3 conv => 5x5 receptive field

Sampler that restricts data loading to a subset of the dataset. It is especially useful in conjunction with :class:`torch.nn.parallel.Distribu

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

Apply a user-defined lambda as a transform. Args: lambd (function): Lambda/function to be used for transform.

Subclass torch's LayerNorm to handle fp16.

MNASNet, as described in https://arxiv.org/pdf/1807.11626.pdf. This implements the B1 variant of the model. >>> model = MNASNet(1000, 1.0)

A MaskedConv2d which inherits the official Conv2d. The masked forward doesn't implement the backward function and only supports the stride pa

r""" Image to Patch Embedding Args: img_size (int): Image size. Default: 224. patch_size (int): Patch token size. Default: 4.

Generates semantic training target only for Panoptic-DeepLab (no instance). Annotation is assumed to have Cityscapes format. Arguments:

Generates semantic training target only for Panoptic-DeepLab (no instance). Annotation is assumed to have Cityscapes format. Arguments:

Sampler that restricts data loading to a subset of the dataset. It is especially useful in conjunction with :class:`torch.nn.parallel.Distrib

Sampler that restricts data loading to a subset of the dataset. It is especially useful in conjunction with :class:`torch.nn.parallel.Distrib

r"""Samples elements randomly from a given list of indices, without replacement. Arguments: indices (sequence): a sequence of indices

Vision Transformer with support for patch or hybrid CNN input stage and image relative position encoding

Vision Transformer with support for patch or hybrid CNN input stage and image relative position encoding

Cell for augmentation Each edge is discrete.

Composes several transforms together. Args: transforms (list of ``Transform`` objects): list of transforms to compose. Example:

Conditionally Parameterized Convolution Inspired by: https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/condconv/condconv

Inverted residual block w/ CondConv routing