|
""" |
|
File: model.py |
|
Author: Elena Ryumina and Dmitry Ryumin |
|
Description: This module provides model architectures. |
|
License: MIT License |
|
""" |
|
|
|
import torch |
|
import torch.nn as nn |
|
import torch.nn.functional as F |
|
import math |
|
|
|
class Bottleneck(nn.Module): |
|
expansion = 4 |
|
def __init__(self, in_channels, out_channels, i_downsample=None, stride=1): |
|
super(Bottleneck, self).__init__() |
|
|
|
self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=stride, padding=0, bias=False) |
|
self.batch_norm1 = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0.99) |
|
|
|
self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding='same', bias=False) |
|
self.batch_norm2 = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0.99) |
|
|
|
self.conv3 = nn.Conv2d(out_channels, out_channels*self.expansion, kernel_size=1, stride=1, padding=0, bias=False) |
|
self.batch_norm3 = nn.BatchNorm2d(out_channels*self.expansion, eps=0.001, momentum=0.99) |
|
|
|
self.i_downsample = i_downsample |
|
self.stride = stride |
|
self.relu = nn.ReLU() |
|
|
|
def forward(self, x): |
|
identity = x.clone() |
|
x = self.relu(self.batch_norm1(self.conv1(x))) |
|
|
|
x = self.relu(self.batch_norm2(self.conv2(x))) |
|
|
|
x = self.conv3(x) |
|
x = self.batch_norm3(x) |
|
|
|
|
|
if self.i_downsample is not None: |
|
identity = self.i_downsample(identity) |
|
|
|
x+=identity |
|
x=self.relu(x) |
|
|
|
return x |
|
|
|
class Conv2dSame(torch.nn.Conv2d): |
|
|
|
def calc_same_pad(self, i: int, k: int, s: int, d: int) -> int: |
|
return max((math.ceil(i / s) - 1) * s + (k - 1) * d + 1 - i, 0) |
|
|
|
def forward(self, x: torch.Tensor) -> torch.Tensor: |
|
ih, iw = x.size()[-2:] |
|
|
|
pad_h = self.calc_same_pad(i=ih, k=self.kernel_size[0], s=self.stride[0], d=self.dilation[0]) |
|
pad_w = self.calc_same_pad(i=iw, k=self.kernel_size[1], s=self.stride[1], d=self.dilation[1]) |
|
|
|
if pad_h > 0 or pad_w > 0: |
|
x = F.pad( |
|
x, [pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2] |
|
) |
|
return F.conv2d( |
|
x, |
|
self.weight, |
|
self.bias, |
|
self.stride, |
|
self.padding, |
|
self.dilation, |
|
self.groups, |
|
) |
|
|
|
class ResNet(nn.Module): |
|
def __init__(self, ResBlock, layer_list, num_classes, num_channels=3): |
|
super(ResNet, self).__init__() |
|
self.in_channels = 64 |
|
|
|
self.conv_layer_s2_same = Conv2dSame(num_channels, 64, 7, stride=2, groups=1, bias=False) |
|
self.batch_norm1 = nn.BatchNorm2d(64, eps=0.001, momentum=0.99) |
|
self.relu = nn.ReLU() |
|
self.max_pool = nn.MaxPool2d(kernel_size = 3, stride=2) |
|
|
|
self.layer1 = self._make_layer(ResBlock, layer_list[0], planes=64, stride=1) |
|
self.layer2 = self._make_layer(ResBlock, layer_list[1], planes=128, stride=2) |
|
self.layer3 = self._make_layer(ResBlock, layer_list[2], planes=256, stride=2) |
|
self.layer4 = self._make_layer(ResBlock, layer_list[3], planes=512, stride=2) |
|
|
|
self.avgpool = nn.AdaptiveAvgPool2d((1,1)) |
|
self.fc1 = nn.Linear(512*ResBlock.expansion, 512) |
|
self.relu1 = nn.ReLU() |
|
self.fc2 = nn.Linear(512, num_classes) |
|
|
|
def extract_features(self, x): |
|
x = self.relu(self.batch_norm1(self.conv_layer_s2_same(x))) |
|
x = self.max_pool(x) |
|
|
|
x = self.layer1(x) |
|
x = self.layer2(x) |
|
x = self.layer3(x) |
|
x = self.layer4(x) |
|
|
|
x = self.avgpool(x) |
|
x = x.reshape(x.shape[0], -1) |
|
x = self.fc1(x) |
|
return x |
|
|
|
def forward(self, x): |
|
x = self.extract_features(x) |
|
x = self.relu1(x) |
|
x = self.fc2(x) |
|
return x |
|
|
|
def _make_layer(self, ResBlock, blocks, planes, stride=1): |
|
ii_downsample = None |
|
layers = [] |
|
|
|
if stride != 1 or self.in_channels != planes*ResBlock.expansion: |
|
ii_downsample = nn.Sequential( |
|
nn.Conv2d(self.in_channels, planes*ResBlock.expansion, kernel_size=1, stride=stride, bias=False, padding=0), |
|
nn.BatchNorm2d(planes*ResBlock.expansion, eps=0.001, momentum=0.99) |
|
) |
|
|
|
layers.append(ResBlock(self.in_channels, planes, i_downsample=ii_downsample, stride=stride)) |
|
self.in_channels = planes*ResBlock.expansion |
|
|
|
for i in range(blocks-1): |
|
layers.append(ResBlock(self.in_channels, planes)) |
|
|
|
return nn.Sequential(*layers) |
|
|
|
def ResNet50(num_classes, channels=3): |
|
return ResNet(Bottleneck, [3,4,6,3], num_classes, channels) |
|
|
|
|
|
class LSTMPyTorch(nn.Module): |
|
def __init__(self): |
|
super(LSTMPyTorch, self).__init__() |
|
|
|
self.lstm1 = nn.LSTM(input_size=512, hidden_size=512, batch_first=True, bidirectional=False) |
|
self.lstm2 = nn.LSTM(input_size=512, hidden_size=256, batch_first=True, bidirectional=False) |
|
self.fc = nn.Linear(256, 7) |
|
self.softmax = nn.Softmax(dim=1) |
|
|
|
def forward(self, x): |
|
x, _ = self.lstm1(x) |
|
x, _ = self.lstm2(x) |
|
x = self.fc(x[:, -1, :]) |
|
x = self.softmax(x) |
|
return x |
