Spaces:
Runtime error
Runtime error
File size: 7,285 Bytes
fc3814c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 |
"""
Video Face Manipulation Detection Through Ensemble of CNNs
Image and Sound Processing Lab - Politecnico di Milano
Nicolò Bonettini
Edoardo Daniele Cannas
Sara Mandelli
Luca Bondi
Paolo Bestagini
"""
from collections import OrderedDict
import torch
from efficientnet_pytorch import EfficientNet
from torch import nn as nn
from torch.nn import functional as F
from torchvision import transforms
from . import externals
"""
Feature Extractor
"""
class FeatureExtractor(nn.Module):
"""
Abstract class to be extended when supporting features extraction.
It also provides standard normalized and parameters
"""
def features(self, x: torch.Tensor) -> torch.Tensor:
raise NotImplementedError
def get_trainable_parameters(self):
return self.parameters()
@staticmethod
def get_normalizer():
return transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
"""
EfficientNet
"""
class EfficientNetGen(FeatureExtractor):
def __init__(self, model: str):
super(EfficientNetGen, self).__init__()
self.efficientnet = EfficientNet.from_pretrained(model)
self.classifier = nn.Linear(self.efficientnet._conv_head.out_channels, 1)
del self.efficientnet._fc
def features(self, x: torch.Tensor) -> torch.Tensor:
x = self.efficientnet.extract_features(x)
x = self.efficientnet._avg_pooling(x)
x = x.flatten(start_dim=1)
return x
def forward(self, x):
x = self.features(x)
x = self.efficientnet._dropout(x)
x = self.classifier(x)
return x
class EfficientNetB4(EfficientNetGen):
def __init__(self):
super(EfficientNetB4, self).__init__(model='efficientnet-b4')
"""
EfficientNetAutoAtt
"""
class EfficientNetAutoAtt(EfficientNet):
def init_att(self, model: str, width: int):
"""
Initialize attention
:param model: efficientnet-bx, x \in {0,..,7}
:param depth: attention width
:return:
"""
if model == 'efficientnet-b4':
self.att_block_idx = 9
if width == 0:
self.attconv = nn.Conv2d(kernel_size=1, in_channels=56, out_channels=1)
else:
attconv_layers = []
for i in range(width):
attconv_layers.append(
('conv{:d}'.format(i), nn.Conv2d(kernel_size=3, padding=1, in_channels=56, out_channels=56)))
attconv_layers.append(
('relu{:d}'.format(i), nn.ReLU(inplace=True)))
attconv_layers.append(('conv_out', nn.Conv2d(kernel_size=1, in_channels=56, out_channels=1)))
self.attconv = nn.Sequential(OrderedDict(attconv_layers))
else:
raise ValueError('Model not valid: {}'.format(model))
def get_attention(self, x: torch.Tensor) -> torch.Tensor:
# Placeholder
att = None
# Stem
x = self._swish(self._bn0(self._conv_stem(x)))
# Blocks
for idx, block in enumerate(self._blocks):
drop_connect_rate = self._global_params.drop_connect_rate
if drop_connect_rate:
drop_connect_rate *= float(idx) / len(self._blocks)
x = block(x, drop_connect_rate=drop_connect_rate)
if idx == self.att_block_idx:
att = torch.sigmoid(self.attconv(x))
break
return att
def extract_features(self, x: torch.Tensor) -> torch.Tensor:
# Stem
x = self._swish(self._bn0(self._conv_stem(x)))
# Blocks
for idx, block in enumerate(self._blocks):
drop_connect_rate = self._global_params.drop_connect_rate
if drop_connect_rate:
drop_connect_rate *= float(idx) / len(self._blocks)
x = block(x, drop_connect_rate=drop_connect_rate)
if idx == self.att_block_idx:
att = torch.sigmoid(self.attconv(x))
x = x * att
# Head
x = self._swish(self._bn1(self._conv_head(x)))
return x
class EfficientNetGenAutoAtt(FeatureExtractor):
def __init__(self, model: str, width: int):
super(EfficientNetGenAutoAtt, self).__init__()
self.efficientnet = EfficientNetAutoAtt.from_pretrained(model)
self.efficientnet.init_att(model, width)
self.classifier = nn.Linear(self.efficientnet._conv_head.out_channels, 1)
del self.efficientnet._fc
def features(self, x: torch.Tensor) -> torch.Tensor:
x = self.efficientnet.extract_features(x)
x = self.efficientnet._avg_pooling(x)
x = x.flatten(start_dim=1)
return x
def forward(self, x):
x = self.features(x)
x = self.efficientnet._dropout(x)
x = self.classifier(x)
return x
def get_attention(self, x: torch.Tensor) -> torch.Tensor:
return self.efficientnet.get_attention(x)
class EfficientNetAutoAttB4(EfficientNetGenAutoAtt):
def __init__(self):
super(EfficientNetAutoAttB4, self).__init__(model='efficientnet-b4', width=0)
"""
Xception
"""
class Xception(FeatureExtractor):
def __init__(self):
super(Xception, self).__init__()
self.xception = externals.xception()
self.xception.last_linear = nn.Linear(2048, 1)
def features(self, x: torch.Tensor) -> torch.Tensor:
x = self.xception.features(x)
x = nn.ReLU(inplace=True)(x)
x = F.adaptive_avg_pool2d(x, (1, 1))
x = x.view(x.size(0), -1)
return x
def forward(self, x: torch.Tensor) -> torch.Tensor:
return self.xception.forward(x)
"""
Siamese tuning
"""
class SiameseTuning(FeatureExtractor):
def __init__(self, feat_ext: FeatureExtractor, num_feat: int, lastonly: bool = True):
super(SiameseTuning, self).__init__()
self.feat_ext = feat_ext()
if not hasattr(self.feat_ext, 'features'):
raise NotImplementedError('The provided feature extractor needs to provide a features() method')
self.lastonly = lastonly
self.classifier = nn.Sequential(
nn.BatchNorm1d(num_features=num_feat),
nn.Linear(in_features=num_feat, out_features=1),
)
def features(self, x):
x = self.feat_ext.features(x)
return x
def forward(self, x: torch.Tensor) -> torch.Tensor:
if self.lastonly:
with torch.no_grad():
x = self.features(x)
else:
x = self.features(x)
x = self.classifier(x)
return x
def get_trainable_parameters(self):
if self.lastonly:
return self.classifier.parameters()
else:
return self.parameters()
class EfficientNetB4ST(SiameseTuning):
def __init__(self):
super(EfficientNetB4ST, self).__init__(feat_ext=EfficientNetB4, num_feat=1792, lastonly=True)
class EfficientNetAutoAttB4ST(SiameseTuning):
def __init__(self):
super(EfficientNetAutoAttB4ST, self).__init__(feat_ext=EfficientNetAutoAttB4, num_feat=1792, lastonly=True)
class XceptionST(SiameseTuning):
def __init__(self):
super(XceptionST, self).__init__(feat_ext=Xception, num_feat=2048, lastonly=True)
|