Spaces:
Sleeping
Sleeping
| import torch | |
| import torch.nn as nn | |
| class ResNeXtBottleneck(nn.Module): | |
| expansion = 2 | |
| def __init__(self, inplanes, planes, stride=1, downsample=None, base_width=2, cardinality=32): | |
| super(ResNeXtBottleneck, self).__init__() | |
| D = int(planes * (base_width / 64.)) * cardinality # = 1024 for planes=1024 | |
| C = cardinality | |
| self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False) | |
| self.bn1 = nn.BatchNorm2d(planes) | |
| self.conv2 = nn.Conv2d(planes, D, kernel_size=3, stride=stride, | |
| padding=1, groups=C, bias=False) | |
| self.bn2 = nn.BatchNorm2d(D) | |
| self.conv3 = nn.Conv2d(D, planes * self.expansion, kernel_size=1, bias=False) | |
| self.bn3 = nn.BatchNorm2d(planes * self.expansion) | |
| self.relu = nn.ReLU(inplace=True) | |
| self.downsample = downsample | |
| self.stride = stride | |
| def forward(self, x): | |
| identity = x | |
| out = self.relu(self.bn1(self.conv1(x))) | |
| out = self.relu(self.bn2(self.conv2(out))) | |
| out = self.bn3(self.conv3(out)) | |
| if self.downsample is not None: | |
| identity = self.downsample(x) | |
| out += identity | |
| out = self.relu(out) | |
| return out | |
| class DeepfakeClassifier(nn.Module): | |
| def __init__(self): | |
| super(DeepfakeClassifier, self).__init__() | |
| self.model = nn.Sequential( | |
| nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False), | |
| nn.BatchNorm2d(64), | |
| nn.ReLU(inplace=True), | |
| nn.MaxPool2d(kernel_size=3, stride=2, padding=1), | |
| self._make_layer(ResNeXtBottleneck, 128, 3, stride=1), # 128*2 = 256 | |
| self._make_layer(ResNeXtBottleneck, 256, 4, stride=2), # 256*2 = 512 | |
| self._make_layer(ResNeXtBottleneck, 512, 6, stride=2), # 512*2 = 1024 | |
| self._make_layer(ResNeXtBottleneck, 1024, 3, stride=2) # 1024*2 = 2048 | |
| ) | |
| self.lstm = nn.LSTM(input_size=2048, hidden_size=2048, batch_first=True, bias=True) | |
| self.linear1 = nn.Linear(2048, 2) | |
| def _make_layer(self, block, planes, blocks, stride=1): | |
| inplanes = self._get_inplanes() | |
| downsample = None | |
| if stride != 1 or inplanes != planes * block.expansion: | |
| downsample = nn.Sequential( | |
| nn.Conv2d(inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False), | |
| nn.BatchNorm2d(planes * block.expansion), | |
| ) | |
| layers = [] | |
| layers.append(block(inplanes, planes, stride, downsample)) | |
| self._inplanes = planes * block.expansion | |
| for _ in range(1, blocks): | |
| layers.append(block(self._inplanes, planes)) | |
| return nn.Sequential(*layers) | |
| def _get_inplanes(self): | |
| if not hasattr(self, '_inplanes'): | |
| self._inplanes = 64 | |
| return self._inplanes | |
| def forward(self, x): | |
| if x.ndim == 4: | |
| # Single image input: (batch, channels, height, width) | |
| x = self.model(x) | |
| x = nn.functional.adaptive_avg_pool2d(x, (1, 1)) | |
| x = x.view(x.size(0), -1) | |
| return self.linear1(x) # No LSTM for images | |
| elif x.ndim == 5: | |
| # Video input: (batch, frames, channels, height, width) | |
| batch, frames, c, h, w = x.shape | |
| x = x.view(-1, c, h, w) | |
| x = self.model(x) | |
| x = nn.functional.adaptive_avg_pool2d(x, (1, 1)) | |
| x = x.view(batch, frames, -1) | |
| x, _ = self.lstm(x) | |
| x = x[:, -1, :] | |
| return self.linear1(x) | |
| else: | |
| raise ValueError(f"Unexpected input shape: {x.shape}") | |