kernel_id
int64
24.2k
23.3M
prompt
stringlengths
8
1.85M
completetion
stringlengths
1
182k
comp_name
stringlengths
5
57
8,522,656
train["fold"] = -1 for fold_id,(trn_idx, val_idx)in enumerate(train_val_indexs): train.loc[val_idx, "fold"] = fold_id train.groupby("fold")[CLASSES].sum()<train_model>
package_path = '.. /input/kha-efficientnet/EfficientNet-PyTorch/' sys.path.append(package_path)
Deepfake Detection Challenge
8,522,656
def resize_images(img_id, input_dir, output_dir, resize_to=(640, 640), ext="png"): img_path = input_dir / f"{img_id}.jpg" save_path = output_dir / f"{img_id}.{ext}" img = cv2.imread(str(img_path), cv2.IMREAD_GRAYSCALE) img = cv2.resize(img, resize_to) cv2.imwrite(str(save_path), img,) TEST_RESIZED = TMP / "test_{0}x{1}".format(*IMAGE_SIZE) TEST_RESIZED.mkdir(exist_ok=True) TEST_RESIZED _ = Parallel(n_jobs=2, verbose=5 )([ delayed(resize_images )(img_id, TEST, TEST_RESIZED, IMAGE_SIZE, "png") for img_id in smpl_sub.StudyInstanceUID.values ] )<choose_model_class>
%%capture !pip install /kaggle/input/khafacenet/facenet_pytorch-2.2.7-py3-none-any.whl !pip install /kaggle/input/imutils/imutils-0.5.3
Deepfake Detection Challenge
8,522,656
def get_activation(activ_name: str="relu"): act_dict = { "relu": nn.ReLU(inplace=True), "tanh": nn.Tanh() , "sigmoid": nn.Sigmoid() , "identity": nn.Identity() } if activ_name in act_dict: return act_dict[activ_name] else: raise NotImplementedError class Conv2dBNActiv(nn.Module): def __init__( self, in_channels: int, out_channels: int, kernel_size: int, stride: int=1, padding: int=0, bias: bool=False, use_bn: bool=True, activ: str="relu" ): super(Conv2dBNActiv, self ).__init__() layers = [] layers.append(nn.Conv2d( in_channels, out_channels, kernel_size, stride, padding, bias=bias)) if use_bn: layers.append(nn.BatchNorm2d(out_channels)) layers.append(get_activation(activ)) self.layers = nn.Sequential(*layers) def forward(self, x): return self.layers(x) class SSEBlock(nn.Module): def __init__(self, in_channels: int): super(SSEBlock, self ).__init__() self.channel_squeeze = nn.Conv2d( in_channels=in_channels, out_channels=1, kernel_size=1, stride=1, padding=0, bias=False) self.sigmoid = nn.Sigmoid() def forward(self, x): h = self.sigmoid(self.channel_squeeze(x)) return x * h class SpatialAttentionBlock(nn.Module): def __init__( self, in_channels: int, out_channels_list: tp.List[int], ): super(SpatialAttentionBlock, self ).__init__() self.n_layers = len(out_channels_list) channels_list = [in_channels] + out_channels_list assert self.n_layers > 0 assert channels_list[-1] == 1 for i in range(self.n_layers - 1): in_chs, out_chs = channels_list[i: i + 2] layer = Conv2dBNActiv(in_chs, out_chs, 3, 1, 1, activ="relu") setattr(self, f"conv{i + 1}", layer) in_chs, out_chs = channels_list[-2:] layer = Conv2dBNActiv(in_chs, out_chs, 3, 1, 1, activ="sigmoid") setattr(self, f"conv{self.n_layers}", layer) def forward(self, x): h = x for i in range(self.n_layers): h = getattr(self, f"conv{i + 1}" )(h) h = h * x return h<choose_model_class>
Deepfake Detection Challenge
8,522,656
class MultiHeadResNet200D(nn.Module): def __init__( self, out_dims_head: tp.List[int]=[3, 4, 3, 1], pretrained=False ): self.base_name = "resnet200d_320" self.n_heads = len(out_dims_head) super(MultiHeadResNet200D, self ).__init__() base_model = timm.create_model( self.base_name, num_classes=sum(out_dims_head), pretrained=False) in_features = base_model.num_features if pretrained: pretrained_model_path = '.. /input/startingpointschestx/resnet200d_320_chestx.pth' state_dict = dict() for k, v in torch.load(pretrained_model_path, map_location='cpu')["model"].items() : if k[:6] == "model.": k = k.replace("model.", "") state_dict[k] = v base_model.load_state_dict(state_dict) base_model.reset_classifier(0, '') self.backbone = base_model for i, out_dim in enumerate(out_dims_head): layer_name = f"head_{i}" layer = nn.Sequential( SpatialAttentionBlock(in_features, [64, 32, 16, 1]), nn.AdaptiveAvgPool2d(output_size=1), nn.Flatten(start_dim=1), nn.Linear(in_features, in_features), nn.ReLU(inplace=True), nn.Dropout(0.5), nn.Linear(in_features, out_dim)) setattr(self, layer_name, layer) def forward(self, x): h = self.backbone(x) hs = [ getattr(self, f"head_{i}" )(h)for i in range(self.n_heads)] y = torch.cat(hs, axis=1) return y m = MultiHeadResNet200D([3, 4, 3, 1], False) m = m.eval() x = torch.rand(1, 3, 256, 256) with torch.no_grad() : y = m(x) print("[forward test]") print("input:\t{} output:\t{}".format(x.shape, y.shape)) del m; del x; del y gc.collect()<categorify>
Deepfake Detection Challenge
8,522,656
class LabeledImageDataset(data.Dataset): def __init__( self, file_list: tp.List[ tp.Tuple[tp.Union[str, Path], tp.Union[int, float, np.ndarray]]], transform_list: tp.List[tp.Dict], ): self.file_list = file_list self.transform = ImageTransformForCls(transform_list) def __len__(self): return len(self.file_list) def __getitem__(self, index): img_path, label = self.file_list[index] img = self._read_image_as_array(img_path) img, label = self.transform(( img, label)) return img, label def _read_image_as_array(self, path: str): img_arr = cv2.imread(str(path)) img_arr = cv2.cvtColor(img_arr, cv2.COLOR_BGR2RGB) return img_arr<create_dataframe>
%matplotlib inline device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
Deepfake Detection Challenge
8,522,656
def get_dataloaders_for_inference( file_list: tp.List[tp.List], batch_size=64, ): dataset = LabeledImageDataset( file_list, transform_list=[ ["Normalize", { "always_apply": True, "max_pixel_value": 255.0, "mean": ["0.4887381077884414"], "std": ["0.23064819430546407"]}], ["ToTensorV2", {"always_apply": True}], ]) loader = data.DataLoader( dataset, batch_size=batch_size, shuffle=False, num_workers=2, pin_memory=True, drop_last=False) return loader<categorify>
def conv_bn(inp, oup, stride, conv_layer=nn.Conv2d, norm_layer=nn.BatchNorm2d, nlin_layer=nn.ReLU): return nn.Sequential( conv_layer(inp, oup, 3, stride, 1, bias=False), norm_layer(oup), nlin_layer(inplace=True) ) def conv_1x1_bn(inp, oup, conv_layer=nn.Conv2d, norm_layer=nn.BatchNorm2d, nlin_layer=nn.ReLU): return nn.Sequential( conv_layer(inp, oup, 1, 1, 0, bias=False), norm_layer(oup), nlin_layer(inplace=True) ) class Hswish(nn.Module): def __init__(self, inplace=True): super(Hswish, self ).__init__() self.inplace = inplace def forward(self, x): return x * F.relu6(x + 3., inplace=self.inplace)/ 6. class Hsigmoid(nn.Module): def __init__(self, inplace=True): super(Hsigmoid, self ).__init__() self.inplace = inplace def forward(self, x): return F.relu6(x + 3., inplace=self.inplace)/ 6. class SEModule(nn.Module): def __init__(self, channel, reduction=4): super(SEModule, self ).__init__() self.avg_pool = nn.AdaptiveAvgPool2d(1) self.fc = nn.Sequential( nn.Linear(channel, channel // reduction, bias=False), nn.ReLU(inplace=True), nn.Linear(channel // reduction, channel, bias=False), Hsigmoid() ) def forward(self, x): b, c, _, _ = x.size() y = self.avg_pool(x ).view(b, c) y = self.fc(y ).view(b, c, 1, 1) return x * y.expand_as(x) class Identity(nn.Module): def __init__(self, channel): super(Identity, self ).__init__() def forward(self, x): return x def make_divisible(x, divisible_by=8): return int(np.ceil(x * 1./ divisible_by)* divisible_by) class MobileBottleneck(nn.Module): def __init__(self, inp, oup, kernel, stride, exp, se=False, nl='RE'): super(MobileBottleneck, self ).__init__() assert stride in [1, 2] assert kernel in [3, 5] padding =(kernel - 1)// 2 self.use_res_connect = stride == 1 and inp == oup conv_layer = nn.Conv2d norm_layer = nn.BatchNorm2d if nl == 'RE': nlin_layer = nn.ReLU elif nl == 'HS': nlin_layer = Hswish else: raise NotImplementedError if se: SELayer = SEModule else: SELayer = Identity self.conv = nn.Sequential( conv_layer(inp, exp, 1, 1, 0, bias=False), norm_layer(exp), nlin_layer(inplace=True), conv_layer(exp, exp, kernel, stride, padding, groups=exp, bias=False), norm_layer(exp), SELayer(exp), nlin_layer(inplace=True), conv_layer(exp, oup, 1, 1, 0, bias=False), norm_layer(oup), ) def forward(self, x): if self.use_res_connect: return x + self.conv(x) else: return self.conv(x) class MobileNetV3(nn.Module): def __init__(self, n_class=1000, input_size=224, dropout=0.8, mode='small', width_mult=1.0): super(MobileNetV3, self ).__init__() input_channel = 16 last_channel = 1280 if mode == 'large': mobile_setting = [ [3, 16, 16, False, 'RE', 1], [3, 64, 24, False, 'RE', 2], [3, 72, 24, False, 'RE', 1], [5, 72, 40, True, 'RE', 2], [5, 120, 40, True, 'RE', 1], [5, 120, 40, True, 'RE', 1], [3, 240, 80, False, 'HS', 2], [3, 200, 80, False, 'HS', 1], [3, 184, 80, False, 'HS', 1], [3, 184, 80, False, 'HS', 1], [3, 480, 112, True, 'HS', 1], [3, 672, 112, True, 'HS', 1], [5, 672, 160, True, 'HS', 2], [5, 960, 160, True, 'HS', 1], [5, 960, 160, True, 'HS', 1], ] elif mode == 'small': mobile_setting = [ [3, 16, 16, True, 'RE', 2], [3, 72, 24, False, 'RE', 2], [3, 88, 24, False, 'RE', 1], [5, 96, 40, True, 'HS', 2], [5, 240, 40, True, 'HS', 1], [5, 240, 40, True, 'HS', 1], [5, 120, 48, True, 'HS', 1], [5, 144, 48, True, 'HS', 1], [5, 288, 96, True, 'HS', 2], [5, 576, 96, True, 'HS', 1], [5, 576, 96, True, 'HS', 1], ] else: raise NotImplementedError assert input_size % 32 == 0 last_channel = make_divisible(last_channel * width_mult)if width_mult > 1.0 else last_channel self.features = [conv_bn(3, input_channel, 2, nlin_layer=Hswish)] self.classifier = [] for k, exp, c, se, nl, s in mobile_setting: output_channel = make_divisible(c * width_mult) exp_channel = make_divisible(exp * width_mult) self.features.append(MobileBottleneck(input_channel, output_channel, k, s, exp_channel, se, nl)) input_channel = output_channel if mode == 'large': last_conv = make_divisible(960 * width_mult) self.features.append(conv_1x1_bn(input_channel, last_conv, nlin_layer=Hswish)) self.features.append(nn.AdaptiveAvgPool2d(1)) self.features.append(nn.Conv2d(last_conv, last_channel, 1, 1, 0)) self.features.append(Hswish(inplace=True)) elif mode == 'small': last_conv = make_divisible(576 * width_mult) self.features.append(conv_1x1_bn(input_channel, last_conv, nlin_layer=Hswish)) self.features.append(nn.AdaptiveAvgPool2d(1)) self.features.append(nn.Conv2d(last_conv, last_channel, 1, 1, 0)) self.features.append(Hswish(inplace=True)) else: raise NotImplementedError self.features = nn.Sequential(*self.features) self.classifier = nn.Sequential( nn.Dropout(p=dropout), nn.Linear(last_channel, n_class), ) self._initialize_weights() def forward(self, x): x = self.features(x) x = x.mean(3 ).mean(2) x = self.classifier(x) return x def _initialize_weights(self): for m in self.modules() : if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode='fan_out') if m.bias is not None: nn.init.zeros_(m.bias) elif isinstance(m, nn.BatchNorm2d): nn.init.ones_(m.weight) nn.init.zeros_(m.bias) elif isinstance(m, nn.Linear): nn.init.normal_(m.weight, 0, 0.01) if m.bias is not None: nn.init.zeros_(m.bias) def mobilenetv3(pretrained=False, **kwargs): model = MobileNetV3(**kwargs) if pretrained: state_dict = torch.load(CFG.pretrained) model.load_state_dict(state_dict, strict=True) return model
Deepfake Detection Challenge
8,522,656
class ImageTransformBase: def __init__(self, data_augmentations: tp.List[tp.Tuple[str, tp.Dict]]): augmentations_list = [ self._get_augmentation(aug_name )(**params) for aug_name, params in data_augmentations] self.data_aug = albumentations.Compose(augmentations_list) def __call__(self, pair: tp.Tuple[np.ndarray])-> tp.Tuple[np.ndarray]: raise NotImplementedError def _get_augmentation(self, aug_name: str)-> tp.Tuple[ImageOnlyTransform, DualTransform]: if hasattr(albumentations, aug_name): return getattr(albumentations, aug_name) else: return eval(aug_name) class ImageTransformForCls(ImageTransformBase): def __init__(self, data_augmentations: tp.List[tp.Tuple[str, tp.Dict]]): super(ImageTransformForCls, self ).__init__(data_augmentations) def __call__(self, in_arrs: tp.Tuple[np.ndarray])-> tp.Tuple[np.ndarray]: img, label = in_arrs augmented = self.data_aug(image=img) img = augmented["image"] return img, label<load_pretrained>
net = mobilenetv3(mode='small', pretrained=False) net.classifier[1] = torch.nn.Linear(in_features=1280, out_features=1) net = net.to(device) state_dict = torch.load(CHECKPOINT) net.load_state_dict(state_dict) net.cuda() net.eval()
Deepfake Detection Challenge
8,522,656
def load_setting_file(path: str): with open(path)as f: settings = yaml.safe_load(f) return settings def set_random_seed(seed: int = 42, deterministic: bool = False): random.seed(seed) np.random.seed(seed) os.environ["PYTHONHASHSEED"] = str(seed) torch.manual_seed(seed) torch.cuda.manual_seed(seed) torch.backends.cudnn.deterministic = deterministic def run_inference_loop(stgs, model, loader, device): model.to(device) model.eval() pred_list = [] with torch.no_grad() : for x, t in tqdm(loader): y_preds1 = model(x.to(device)) y_preds2 = model(x.flip(-1 ).to(device)) y_preds =(y_preds1.sigmoid().to('cpu' ).numpy() *0.6 + y_preds2.sigmoid().to('cpu' ).numpy() *0.4) pred_list.append(y_preds) pred_arr = np.concatenate(pred_list) del pred_list return pred_arr<set_options>
net2 = mobilenetv3(mode='small', pretrained=False) net2.classifier[1] = torch.nn.Linear(in_features=1280, out_features=1) net2 = net2.to(device) state_dict = torch.load(CHECKPOINT2) net2.load_state_dict(state_dict) net2.cuda() net2.eval()
Deepfake Detection Challenge
8,522,656
if not torch.cuda.is_available() : device = torch.device("cpu") else: device = torch.device("cuda") print(device )<load_pretrained>
net3 = EfficientNet.from_name("efficientnet-b0") net3._fc = torch.nn.Linear(in_features=net3._fc.in_features, out_features=1) net3.load_state_dict(torch.load(CHECKPOINT3)) net3.cuda() net3.eval()
Deepfake Detection Challenge
8,522,656
model_dir = TRAINED_MODEL test_dir = TEST_RESIZED test_file_list = [ (test_dir / f"{img_id}.png", [-1] * 11) for img_id in smpl_sub["StudyInstanceUID"].values] test_loader = get_dataloaders_for_inference(test_file_list, batch_size=32) test_preds_arr = np.zeros(( N_FOLD , len(smpl_sub), N_CLASSES)) for fold_id in [0,1,2,3,4]: print(f"[fold {fold_id}]") stgs = load_setting_file(".. /input/ranzer-models/settings.yml") stgs["model"]["params"]["pretrained"] = False model = MultiHeadResNet200D(**stgs["model"]["params"]) model_path = model_dir / f"best_model_fold{fold_id}.pth" print(model_path) model.load_state_dict(torch.load(model_path, map_location=device)) test_pred = run_inference_loop(stgs, model,test_loader, device) test_preds_arr[fold_id] = test_pred del model torch.cuda.empty_cache() gc.collect() <save_to_csv>
net6 = EfficientNet.from_name("efficientnet-b1") net6._fc = torch.nn.Linear(in_features=net6._fc.in_features, out_features=1) net6.load_state_dict(torch.load(CHECKPOINT6)) net6.cuda() net6.eval()
Deepfake Detection Challenge
8,522,656
if CONVERT_TO_RANK: test_preds_arr = test_preds_arr.argsort(axis=1 ).argsort(axis=1) sub = smpl_sub.copy() sub[CLASSES] = test_preds_arr.mean(axis=0) sub.to_csv("submission.csv", index=False )<load_pretrained>
class SeparableConv2d(nn.Module): def __init__(self,in_channels,out_channels,kernel_size=1,stride=1,padding=0,dilation=1,bias=False): super(SeparableConv2d,self ).__init__() self.conv1 = nn.Conv2d(in_channels,in_channels,kernel_size,stride,padding,dilation,groups=in_channels,bias=bias) self.pointwise = nn.Conv2d(in_channels,out_channels,1,1,0,1,1,bias=bias) def forward(self,x): x = self.conv1(x) x = self.pointwise(x) return x class Block(nn.Module): def __init__(self,in_filters,out_filters,reps,strides=1,start_with_relu=True,grow_first=True): super(Block, self ).__init__() if out_filters != in_filters or strides!=1: self.skip = nn.Conv2d(in_filters,out_filters,1,stride=strides, bias=False) self.skipbn = nn.BatchNorm2d(out_filters) else: self.skip=None rep=[] filters=in_filters if grow_first: rep.append(nn.ReLU(inplace=True)) rep.append(SeparableConv2d(in_filters,out_filters,3,stride=1,padding=1,bias=False)) rep.append(nn.BatchNorm2d(out_filters)) filters = out_filters for i in range(reps-1): rep.append(nn.ReLU(inplace=True)) rep.append(SeparableConv2d(filters,filters,3,stride=1,padding=1,bias=False)) rep.append(nn.BatchNorm2d(filters)) if not grow_first: rep.append(nn.ReLU(inplace=True)) rep.append(SeparableConv2d(in_filters,out_filters,3,stride=1,padding=1,bias=False)) rep.append(nn.BatchNorm2d(out_filters)) if not start_with_relu: rep = rep[1:] else: rep[0] = nn.ReLU(inplace=False) if strides != 1: rep.append(nn.MaxPool2d(3,strides,1)) self.rep = nn.Sequential(*rep) def forward(self,inp): x = self.rep(inp) if self.skip is not None: skip = self.skip(inp) skip = self.skipbn(skip) else: skip = inp x+=skip return x class Xception(nn.Module): def __init__(self, num_classes=1000): super(Xception, self ).__init__() self.num_classes = num_classes self.conv1 = nn.Conv2d(3, 32, 3,2, 0, bias=False) self.bn1 = nn.BatchNorm2d(32) self.relu1 = nn.ReLU(inplace=True) self.conv2 = nn.Conv2d(32,64,3,bias=False) self.bn2 = nn.BatchNorm2d(64) self.relu2 = nn.ReLU(inplace=True) self.block1=Block(64,128,2,2,start_with_relu=False,grow_first=True) self.block2=Block(128,256,2,2,start_with_relu=True,grow_first=True) self.block3=Block(256,728,2,2,start_with_relu=True,grow_first=True) self.block4=Block(728,728,3,1,start_with_relu=True,grow_first=True) self.block5=Block(728,728,3,1,start_with_relu=True,grow_first=True) self.block6=Block(728,728,3,1,start_with_relu=True,grow_first=True) self.block7=Block(728,728,3,1,start_with_relu=True,grow_first=True) self.block8=Block(728,728,3,1,start_with_relu=True,grow_first=True) self.block9=Block(728,728,3,1,start_with_relu=True,grow_first=True) self.block10=Block(728,728,3,1,start_with_relu=True,grow_first=True) self.block11=Block(728,728,3,1,start_with_relu=True,grow_first=True) self.block12=Block(728,1024,2,2,start_with_relu=True,grow_first=False) self.conv3 = SeparableConv2d(1024,1536,3,1,1) self.bn3 = nn.BatchNorm2d(1536) self.relu3 = nn.ReLU(inplace=True) self.conv4 = SeparableConv2d(1536,2048,3,1,1) self.bn4 = nn.BatchNorm2d(2048) self.fc = nn.Linear(2048, num_classes) def features(self, input): x = self.conv1(input) x = self.bn1(x) x = self.relu1(x) x = self.conv2(x) x = self.bn2(x) x = self.relu2(x) x = self.block1(x) x = self.block2(x) x = self.block3(x) x = self.block4(x) x = self.block5(x) x = self.block6(x) x = self.block7(x) x = self.block8(x) x = self.block9(x) x = self.block10(x) x = self.block11(x) x = self.block12(x) x = self.conv3(x) x = self.bn3(x) x = self.relu3(x) x = self.conv4(x) x = self.bn4(x) return x def logits(self, features): x = nn.ReLU(inplace=True )(features) x = F.adaptive_avg_pool2d(x,(1, 1)) x = x.view(x.size(0), -1) x = self.last_linear(x) return x def forward(self, input): x = self.features(input) x = self.logits(x) return x def xception(num_classes=1000, pretrained='imagenet'): model = Xception(num_classes=num_classes) if pretrained: settings = pretrained_settings['xception'][pretrained] assert num_classes == settings['num_classes'], \ "num_classes should be {}, but is {}".format(settings['num_classes'], num_classes) model = Xception(num_classes=num_classes) model.load_state_dict(model_zoo.load_url(settings['url'])) model.input_space = settings['input_space'] model.input_size = settings['input_size'] model.input_range = settings['input_range'] model.mean = settings['mean'] model.std = settings['std'] model.last_linear = model.fc del model.fc return model net4 = Xception() net4.last_linear = net4.fc del net4.fc net4.last_linear = nn.Linear(in_features=2048, out_features=1, bias=True) net4.load_state_dict(torch.load(CHECKPOINT4)) net4 = net4.to(device) net4.cuda() net4.eval() net10 = Xception() net10.last_linear = net10.fc del net10.fc net10.last_linear = nn.Linear(in_features=2048, out_features=1, bias=True) net10.load_state_dict(torch.load(CHECKPOINT10)) net10 = net10.to(device) net10.cuda() net10.eval()
Deepfake Detection Challenge
8,522,656
model_dir = TRAINED_MODEL test_dir = TEST_RESIZED test_file_list = [ (test_dir / f"{img_id}.png", [-1] * 11) for img_id in smpl_sub["StudyInstanceUID"].values] test_loader = get_dataloaders_for_inference(test_file_list, batch_size=4) N_FOLD = len([1024]) test_preds_arr = np.zeros(( N_FOLD , len(smpl_sub), N_CLASSES)) for i,fold_id in enumerate([1024]): print(f"[fold {fold_id}]") stgs = load_setting_file(".. /input/ranzer-models/settings.yml") stgs["model"]["params"]["pretrained"] = False model = MultiHeadResNet200D(**stgs["model"]["params"]) model_path = model_dir / f"best_model_fold{fold_id}.pth" print(model_path) model.load_state_dict(torch.load(model_path, map_location=device)) test_pred = run_inference_loop(stgs, model,test_loader, device) test_preds_arr[i] = test_pred del model torch.cuda.empty_cache() gc.collect() <prepare_output>
class MaxPoolPad(nn.Module): def __init__(self): super(MaxPoolPad, self ).__init__() self.pad = nn.ZeroPad2d(( 1, 0, 1, 0)) self.pool = nn.MaxPool2d(3, stride=2, padding=1) def forward(self, x): x = self.pad(x) x = self.pool(x) x = x[:, :, 1:, 1:].contiguous() return x class AvgPoolPad(nn.Module): def __init__(self, stride=2, padding=1): super(AvgPoolPad, self ).__init__() self.pad = nn.ZeroPad2d(( 1, 0, 1, 0)) self.pool = nn.AvgPool2d(3, stride=stride, padding=padding, count_include_pad=False) def forward(self, x): x = self.pad(x) x = self.pool(x) x = x[:, :, 1:, 1:].contiguous() return x class SeparableConv2d(nn.Module): def __init__(self, in_channels, out_channels, dw_kernel, dw_stride, dw_padding, bias=False): super(SeparableConv2d, self ).__init__() self.depthwise_conv2d = nn.Conv2d(in_channels, in_channels, dw_kernel, stride=dw_stride, padding=dw_padding, bias=bias, groups=in_channels) self.pointwise_conv2d = nn.Conv2d(in_channels, out_channels, 1, stride=1, bias=bias) def forward(self, x): x = self.depthwise_conv2d(x) x = self.pointwise_conv2d(x) return x class BranchSeparables(nn.Module): def __init__(self, in_channels, out_channels, kernel_size, stride, padding, name=None, bias=False): super(BranchSeparables, self ).__init__() self.relu = nn.ReLU() self.separable_1 = SeparableConv2d(in_channels, in_channels, kernel_size, stride, padding, bias=bias) self.bn_sep_1 = nn.BatchNorm2d(in_channels, eps=0.001, momentum=0.1, affine=True) self.relu1 = nn.ReLU() self.separable_2 = SeparableConv2d(in_channels, out_channels, kernel_size, 1, padding, bias=bias) self.bn_sep_2 = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0.1, affine=True) self.name = name def forward(self, x): x = self.relu(x) if self.name == 'specific': x = nn.ZeroPad2d(( 1, 0, 1, 0))(x) x = self.separable_1(x) if self.name == 'specific': x = x[:, :, 1:, 1:].contiguous() x = self.bn_sep_1(x) x = self.relu1(x) x = self.separable_2(x) x = self.bn_sep_2(x) return x class BranchSeparablesStem(nn.Module): def __init__(self, in_channels, out_channels, kernel_size, stride, padding, bias=False): super(BranchSeparablesStem, self ).__init__() self.relu = nn.ReLU() self.separable_1 = SeparableConv2d(in_channels, out_channels, kernel_size, stride, padding, bias=bias) self.bn_sep_1 = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0.1, affine=True) self.relu1 = nn.ReLU() self.separable_2 = SeparableConv2d(out_channels, out_channels, kernel_size, 1, padding, bias=bias) self.bn_sep_2 = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0.1, affine=True) def forward(self, x): x = self.relu(x) x = self.separable_1(x) x = self.bn_sep_1(x) x = self.relu1(x) x = self.separable_2(x) x = self.bn_sep_2(x) return x class BranchSeparablesReduction(BranchSeparables): def __init__(self, in_channels, out_channels, kernel_size, stride, padding, z_padding=1, bias=False): BranchSeparables.__init__(self, in_channels, out_channels, kernel_size, stride, padding, bias) self.padding = nn.ZeroPad2d(( z_padding, 0, z_padding, 0)) def forward(self, x): x = self.relu(x) x = self.padding(x) x = self.separable_1(x) x = x[:, :, 1:, 1:].contiguous() x = self.bn_sep_1(x) x = self.relu1(x) x = self.separable_2(x) x = self.bn_sep_2(x) return x class CellStem0(nn.Module): def __init__(self, stem_filters, num_filters=42): super(CellStem0, self ).__init__() self.num_filters = num_filters self.stem_filters = stem_filters self.conv_1x1 = nn.Sequential() self.conv_1x1.add_module('relu', nn.ReLU()) self.conv_1x1.add_module('conv', nn.Conv2d(self.stem_filters, self.num_filters, 1, stride=1, bias=False)) self.conv_1x1.add_module('bn', nn.BatchNorm2d(self.num_filters, eps=0.001, momentum=0.1, affine=True)) self.comb_iter_0_left = BranchSeparables(self.num_filters, self.num_filters, 5, 2, 2) self.comb_iter_0_right = BranchSeparablesStem(self.stem_filters, self.num_filters, 7, 2, 3, bias=False) self.comb_iter_1_left = nn.MaxPool2d(3, stride=2, padding=1) self.comb_iter_1_right = BranchSeparablesStem(self.stem_filters, self.num_filters, 7, 2, 3, bias=False) self.comb_iter_2_left = nn.AvgPool2d(3, stride=2, padding=1, count_include_pad=False) self.comb_iter_2_right = BranchSeparablesStem(self.stem_filters, self.num_filters, 5, 2, 2, bias=False) self.comb_iter_3_right = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_4_left = BranchSeparables(self.num_filters, self.num_filters, 3, 1, 1, bias=False) self.comb_iter_4_right = nn.MaxPool2d(3, stride=2, padding=1) def forward(self, x): x1 = self.conv_1x1(x) x_comb_iter_0_left = self.comb_iter_0_left(x1) x_comb_iter_0_right = self.comb_iter_0_right(x) x_comb_iter_0 = x_comb_iter_0_left + x_comb_iter_0_right x_comb_iter_1_left = self.comb_iter_1_left(x1) x_comb_iter_1_right = self.comb_iter_1_right(x) x_comb_iter_1 = x_comb_iter_1_left + x_comb_iter_1_right x_comb_iter_2_left = self.comb_iter_2_left(x1) x_comb_iter_2_right = self.comb_iter_2_right(x) x_comb_iter_2 = x_comb_iter_2_left + x_comb_iter_2_right x_comb_iter_3_right = self.comb_iter_3_right(x_comb_iter_0) x_comb_iter_3 = x_comb_iter_3_right + x_comb_iter_1 x_comb_iter_4_left = self.comb_iter_4_left(x_comb_iter_0) x_comb_iter_4_right = self.comb_iter_4_right(x1) x_comb_iter_4 = x_comb_iter_4_left + x_comb_iter_4_right x_out = torch.cat([x_comb_iter_1, x_comb_iter_2, x_comb_iter_3, x_comb_iter_4], 1) return x_out class CellStem1(nn.Module): def __init__(self, stem_filters, num_filters): super(CellStem1, self ).__init__() self.num_filters = num_filters self.stem_filters = stem_filters self.conv_1x1 = nn.Sequential() self.conv_1x1.add_module('relu', nn.ReLU()) self.conv_1x1.add_module('conv', nn.Conv2d(2*self.num_filters, self.num_filters, 1, stride=1, bias=False)) self.conv_1x1.add_module('bn', nn.BatchNorm2d(self.num_filters, eps=0.001, momentum=0.1, affine=True)) self.relu = nn.ReLU() self.path_1 = nn.Sequential() self.path_1.add_module('avgpool', nn.AvgPool2d(1, stride=2, count_include_pad=False)) self.path_1.add_module('conv', nn.Conv2d(self.stem_filters, self.num_filters//2, 1, stride=1, bias=False)) self.path_2 = nn.ModuleList() self.path_2.add_module('pad', nn.ZeroPad2d(( 0, 1, 0, 1))) self.path_2.add_module('avgpool', nn.AvgPool2d(1, stride=2, count_include_pad=False)) self.path_2.add_module('conv', nn.Conv2d(self.stem_filters, self.num_filters//2, 1, stride=1, bias=False)) self.final_path_bn = nn.BatchNorm2d(self.num_filters, eps=0.001, momentum=0.1, affine=True) self.comb_iter_0_left = BranchSeparables(self.num_filters, self.num_filters, 5, 2, 2, name='specific', bias=False) self.comb_iter_0_right = BranchSeparables(self.num_filters, self.num_filters, 7, 2, 3, name='specific', bias=False) self.comb_iter_1_left = MaxPoolPad() self.comb_iter_1_right = BranchSeparables(self.num_filters, self.num_filters, 7, 2, 3, name='specific', bias=False) self.comb_iter_2_left = AvgPoolPad() self.comb_iter_2_right = BranchSeparables(self.num_filters, self.num_filters, 5, 2, 2, name='specific', bias=False) self.comb_iter_3_right = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_4_left = BranchSeparables(self.num_filters, self.num_filters, 3, 1, 1, name='specific', bias=False) self.comb_iter_4_right = MaxPoolPad() def forward(self, x_conv0, x_stem_0): x_left = self.conv_1x1(x_stem_0) x_relu = self.relu(x_conv0) x_path1 = self.path_1(x_relu) x_path2 = self.path_2.pad(x_relu) x_path2 = x_path2[:, :, 1:, 1:] x_path2 = self.path_2.avgpool(x_path2) x_path2 = self.path_2.conv(x_path2) x_right = self.final_path_bn(torch.cat([x_path1, x_path2], 1)) x_comb_iter_0_left = self.comb_iter_0_left(x_left) x_comb_iter_0_right = self.comb_iter_0_right(x_right) x_comb_iter_0 = x_comb_iter_0_left + x_comb_iter_0_right x_comb_iter_1_left = self.comb_iter_1_left(x_left) x_comb_iter_1_right = self.comb_iter_1_right(x_right) x_comb_iter_1 = x_comb_iter_1_left + x_comb_iter_1_right x_comb_iter_2_left = self.comb_iter_2_left(x_left) x_comb_iter_2_right = self.comb_iter_2_right(x_right) x_comb_iter_2 = x_comb_iter_2_left + x_comb_iter_2_right x_comb_iter_3_right = self.comb_iter_3_right(x_comb_iter_0) x_comb_iter_3 = x_comb_iter_3_right + x_comb_iter_1 x_comb_iter_4_left = self.comb_iter_4_left(x_comb_iter_0) x_comb_iter_4_right = self.comb_iter_4_right(x_left) x_comb_iter_4 = x_comb_iter_4_left + x_comb_iter_4_right x_out = torch.cat([x_comb_iter_1, x_comb_iter_2, x_comb_iter_3, x_comb_iter_4], 1) return x_out class FirstCell(nn.Module): def __init__(self, in_channels_left, out_channels_left, in_channels_right, out_channels_right): super(FirstCell, self ).__init__() self.conv_1x1 = nn.Sequential() self.conv_1x1.add_module('relu', nn.ReLU()) self.conv_1x1.add_module('conv', nn.Conv2d(in_channels_right, out_channels_right, 1, stride=1, bias=False)) self.conv_1x1.add_module('bn', nn.BatchNorm2d(out_channels_right, eps=0.001, momentum=0.1, affine=True)) self.relu = nn.ReLU() self.path_1 = nn.Sequential() self.path_1.add_module('avgpool', nn.AvgPool2d(1, stride=2, count_include_pad=False)) self.path_1.add_module('conv', nn.Conv2d(in_channels_left, out_channels_left, 1, stride=1, bias=False)) self.path_2 = nn.ModuleList() self.path_2.add_module('pad', nn.ZeroPad2d(( 0, 1, 0, 1))) self.path_2.add_module('avgpool', nn.AvgPool2d(1, stride=2, count_include_pad=False)) self.path_2.add_module('conv', nn.Conv2d(in_channels_left, out_channels_left, 1, stride=1, bias=False)) self.final_path_bn = nn.BatchNorm2d(out_channels_left * 2, eps=0.001, momentum=0.1, affine=True) self.comb_iter_0_left = BranchSeparables(out_channels_right, out_channels_right, 5, 1, 2, bias=False) self.comb_iter_0_right = BranchSeparables(out_channels_right, out_channels_right, 3, 1, 1, bias=False) self.comb_iter_1_left = BranchSeparables(out_channels_right, out_channels_right, 5, 1, 2, bias=False) self.comb_iter_1_right = BranchSeparables(out_channels_right, out_channels_right, 3, 1, 1, bias=False) self.comb_iter_2_left = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_3_left = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_3_right = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_4_left = BranchSeparables(out_channels_right, out_channels_right, 3, 1, 1, bias=False) def forward(self, x, x_prev): x_relu = self.relu(x_prev) x_path1 = self.path_1(x_relu) x_path2 = self.path_2.pad(x_relu) x_path2 = x_path2[:, :, 1:, 1:] x_path2 = self.path_2.avgpool(x_path2) x_path2 = self.path_2.conv(x_path2) x_left = self.final_path_bn(torch.cat([x_path1, x_path2], 1)) x_right = self.conv_1x1(x) x_comb_iter_0_left = self.comb_iter_0_left(x_right) x_comb_iter_0_right = self.comb_iter_0_right(x_left) x_comb_iter_0 = x_comb_iter_0_left + x_comb_iter_0_right x_comb_iter_1_left = self.comb_iter_1_left(x_left) x_comb_iter_1_right = self.comb_iter_1_right(x_left) x_comb_iter_1 = x_comb_iter_1_left + x_comb_iter_1_right x_comb_iter_2_left = self.comb_iter_2_left(x_right) x_comb_iter_2 = x_comb_iter_2_left + x_left x_comb_iter_3_left = self.comb_iter_3_left(x_left) x_comb_iter_3_right = self.comb_iter_3_right(x_left) x_comb_iter_3 = x_comb_iter_3_left + x_comb_iter_3_right x_comb_iter_4_left = self.comb_iter_4_left(x_right) x_comb_iter_4 = x_comb_iter_4_left + x_right x_out = torch.cat([x_left, x_comb_iter_0, x_comb_iter_1, x_comb_iter_2, x_comb_iter_3, x_comb_iter_4], 1) return x_out class NormalCell(nn.Module): def __init__(self, in_channels_left, out_channels_left, in_channels_right, out_channels_right): super(NormalCell, self ).__init__() self.conv_prev_1x1 = nn.Sequential() self.conv_prev_1x1.add_module('relu', nn.ReLU()) self.conv_prev_1x1.add_module('conv', nn.Conv2d(in_channels_left, out_channels_left, 1, stride=1, bias=False)) self.conv_prev_1x1.add_module('bn', nn.BatchNorm2d(out_channels_left, eps=0.001, momentum=0.1, affine=True)) self.conv_1x1 = nn.Sequential() self.conv_1x1.add_module('relu', nn.ReLU()) self.conv_1x1.add_module('conv', nn.Conv2d(in_channels_right, out_channels_right, 1, stride=1, bias=False)) self.conv_1x1.add_module('bn', nn.BatchNorm2d(out_channels_right, eps=0.001, momentum=0.1, affine=True)) self.comb_iter_0_left = BranchSeparables(out_channels_right, out_channels_right, 5, 1, 2, bias=False) self.comb_iter_0_right = BranchSeparables(out_channels_left, out_channels_left, 3, 1, 1, bias=False) self.comb_iter_1_left = BranchSeparables(out_channels_left, out_channels_left, 5, 1, 2, bias=False) self.comb_iter_1_right = BranchSeparables(out_channels_left, out_channels_left, 3, 1, 1, bias=False) self.comb_iter_2_left = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_3_left = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_3_right = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_4_left = BranchSeparables(out_channels_right, out_channels_right, 3, 1, 1, bias=False) def forward(self, x, x_prev): x_left = self.conv_prev_1x1(x_prev) x_right = self.conv_1x1(x) x_comb_iter_0_left = self.comb_iter_0_left(x_right) x_comb_iter_0_right = self.comb_iter_0_right(x_left) x_comb_iter_0 = x_comb_iter_0_left + x_comb_iter_0_right x_comb_iter_1_left = self.comb_iter_1_left(x_left) x_comb_iter_1_right = self.comb_iter_1_right(x_left) x_comb_iter_1 = x_comb_iter_1_left + x_comb_iter_1_right x_comb_iter_2_left = self.comb_iter_2_left(x_right) x_comb_iter_2 = x_comb_iter_2_left + x_left x_comb_iter_3_left = self.comb_iter_3_left(x_left) x_comb_iter_3_right = self.comb_iter_3_right(x_left) x_comb_iter_3 = x_comb_iter_3_left + x_comb_iter_3_right x_comb_iter_4_left = self.comb_iter_4_left(x_right) x_comb_iter_4 = x_comb_iter_4_left + x_right x_out = torch.cat([x_left, x_comb_iter_0, x_comb_iter_1, x_comb_iter_2, x_comb_iter_3, x_comb_iter_4], 1) return x_out class ReductionCell0(nn.Module): def __init__(self, in_channels_left, out_channels_left, in_channels_right, out_channels_right): super(ReductionCell0, self ).__init__() self.conv_prev_1x1 = nn.Sequential() self.conv_prev_1x1.add_module('relu', nn.ReLU()) self.conv_prev_1x1.add_module('conv', nn.Conv2d(in_channels_left, out_channels_left, 1, stride=1, bias=False)) self.conv_prev_1x1.add_module('bn', nn.BatchNorm2d(out_channels_left, eps=0.001, momentum=0.1, affine=True)) self.conv_1x1 = nn.Sequential() self.conv_1x1.add_module('relu', nn.ReLU()) self.conv_1x1.add_module('conv', nn.Conv2d(in_channels_right, out_channels_right, 1, stride=1, bias=False)) self.conv_1x1.add_module('bn', nn.BatchNorm2d(out_channels_right, eps=0.001, momentum=0.1, affine=True)) self.comb_iter_0_left = BranchSeparablesReduction(out_channels_right, out_channels_right, 5, 2, 2, bias=False) self.comb_iter_0_right = BranchSeparablesReduction(out_channels_right, out_channels_right, 7, 2, 3, bias=False) self.comb_iter_1_left = MaxPoolPad() self.comb_iter_1_right = BranchSeparablesReduction(out_channels_right, out_channels_right, 7, 2, 3, bias=False) self.comb_iter_2_left = AvgPoolPad() self.comb_iter_2_right = BranchSeparablesReduction(out_channels_right, out_channels_right, 5, 2, 2, bias=False) self.comb_iter_3_right = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_4_left = BranchSeparablesReduction(out_channels_right, out_channels_right, 3, 1, 1, bias=False) self.comb_iter_4_right = MaxPoolPad() def forward(self, x, x_prev): x_left = self.conv_prev_1x1(x_prev) x_right = self.conv_1x1(x) x_comb_iter_0_left = self.comb_iter_0_left(x_right) x_comb_iter_0_right = self.comb_iter_0_right(x_left) x_comb_iter_0 = x_comb_iter_0_left + x_comb_iter_0_right x_comb_iter_1_left = self.comb_iter_1_left(x_right) x_comb_iter_1_right = self.comb_iter_1_right(x_left) x_comb_iter_1 = x_comb_iter_1_left + x_comb_iter_1_right x_comb_iter_2_left = self.comb_iter_2_left(x_right) x_comb_iter_2_right = self.comb_iter_2_right(x_left) x_comb_iter_2 = x_comb_iter_2_left + x_comb_iter_2_right x_comb_iter_3_right = self.comb_iter_3_right(x_comb_iter_0) x_comb_iter_3 = x_comb_iter_3_right + x_comb_iter_1 x_comb_iter_4_left = self.comb_iter_4_left(x_comb_iter_0) x_comb_iter_4_right = self.comb_iter_4_right(x_right) x_comb_iter_4 = x_comb_iter_4_left + x_comb_iter_4_right x_out = torch.cat([x_comb_iter_1, x_comb_iter_2, x_comb_iter_3, x_comb_iter_4], 1) return x_out class ReductionCell1(nn.Module): def __init__(self, in_channels_left, out_channels_left, in_channels_right, out_channels_right): super(ReductionCell1, self ).__init__() self.conv_prev_1x1 = nn.Sequential() self.conv_prev_1x1.add_module('relu', nn.ReLU()) self.conv_prev_1x1.add_module('conv', nn.Conv2d(in_channels_left, out_channels_left, 1, stride=1, bias=False)) self.conv_prev_1x1.add_module('bn', nn.BatchNorm2d(out_channels_left, eps=0.001, momentum=0.1, affine=True)) self.conv_1x1 = nn.Sequential() self.conv_1x1.add_module('relu', nn.ReLU()) self.conv_1x1.add_module('conv', nn.Conv2d(in_channels_right, out_channels_right, 1, stride=1, bias=False)) self.conv_1x1.add_module('bn', nn.BatchNorm2d(out_channels_right, eps=0.001, momentum=0.1, affine=True)) self.comb_iter_0_left = BranchSeparables(out_channels_right, out_channels_right, 5, 2, 2, name='specific', bias=False) self.comb_iter_0_right = BranchSeparables(out_channels_right, out_channels_right, 7, 2, 3, name='specific', bias=False) self.comb_iter_1_left = MaxPoolPad() self.comb_iter_1_right = BranchSeparables(out_channels_right, out_channels_right, 7, 2, 3, name='specific', bias=False) self.comb_iter_2_left = AvgPoolPad() self.comb_iter_2_right = BranchSeparables(out_channels_right, out_channels_right, 5, 2, 2, name='specific', bias=False) self.comb_iter_3_right = nn.AvgPool2d(3, stride=1, padding=1, count_include_pad=False) self.comb_iter_4_left = BranchSeparables(out_channels_right, out_channels_right, 3, 1, 1, name='specific', bias=False) self.comb_iter_4_right =MaxPoolPad() def forward(self, x, x_prev): x_left = self.conv_prev_1x1(x_prev) x_right = self.conv_1x1(x) x_comb_iter_0_left = self.comb_iter_0_left(x_right) x_comb_iter_0_right = self.comb_iter_0_right(x_left) x_comb_iter_0 = x_comb_iter_0_left + x_comb_iter_0_right x_comb_iter_1_left = self.comb_iter_1_left(x_right) x_comb_iter_1_right = self.comb_iter_1_right(x_left) x_comb_iter_1 = x_comb_iter_1_left + x_comb_iter_1_right x_comb_iter_2_left = self.comb_iter_2_left(x_right) x_comb_iter_2_right = self.comb_iter_2_right(x_left) x_comb_iter_2 = x_comb_iter_2_left + x_comb_iter_2_right x_comb_iter_3_right = self.comb_iter_3_right(x_comb_iter_0) x_comb_iter_3 = x_comb_iter_3_right + x_comb_iter_1 x_comb_iter_4_left = self.comb_iter_4_left(x_comb_iter_0) x_comb_iter_4_right = self.comb_iter_4_right(x_right) x_comb_iter_4 = x_comb_iter_4_left + x_comb_iter_4_right x_out = torch.cat([x_comb_iter_1, x_comb_iter_2, x_comb_iter_3, x_comb_iter_4], 1) return x_out class NASNetAMobile(nn.Module): def __init__(self, num_classes=1000, stem_filters=32, penultimate_filters=1056, filters_multiplier=2): super(NASNetAMobile, self ).__init__() self.num_classes = num_classes self.stem_filters = stem_filters self.penultimate_filters = penultimate_filters self.filters_multiplier = filters_multiplier filters = self.penultimate_filters // 24 self.conv0 = nn.Sequential() self.conv0.add_module('conv', nn.Conv2d(in_channels=3, out_channels=self.stem_filters, kernel_size=3, padding=0, stride=2, bias=False)) self.conv0.add_module('bn', nn.BatchNorm2d(self.stem_filters, eps=0.001, momentum=0.1, affine=True)) self.cell_stem_0 = CellStem0(self.stem_filters, num_filters=filters //(filters_multiplier ** 2)) self.cell_stem_1 = CellStem1(self.stem_filters, num_filters=filters // filters_multiplier) self.cell_0 = FirstCell(in_channels_left=filters, out_channels_left=filters//2, in_channels_right=2*filters, out_channels_right=filters) self.cell_1 = NormalCell(in_channels_left=2*filters, out_channels_left=filters, in_channels_right=6*filters, out_channels_right=filters) self.cell_2 = NormalCell(in_channels_left=6*filters, out_channels_left=filters, in_channels_right=6*filters, out_channels_right=filters) self.cell_3 = NormalCell(in_channels_left=6*filters, out_channels_left=filters, in_channels_right=6*filters, out_channels_right=filters) self.reduction_cell_0 = ReductionCell0(in_channels_left=6*filters, out_channels_left=2*filters, in_channels_right=6*filters, out_channels_right=2*filters) self.cell_6 = FirstCell(in_channels_left=6*filters, out_channels_left=filters, in_channels_right=8*filters, out_channels_right=2*filters) self.cell_7 = NormalCell(in_channels_left=8*filters, out_channels_left=2*filters, in_channels_right=12*filters, out_channels_right=2*filters) self.cell_8 = NormalCell(in_channels_left=12*filters, out_channels_left=2*filters, in_channels_right=12*filters, out_channels_right=2*filters) self.cell_9 = NormalCell(in_channels_left=12*filters, out_channels_left=2*filters, in_channels_right=12*filters, out_channels_right=2*filters) self.reduction_cell_1 = ReductionCell1(in_channels_left=12*filters, out_channels_left=4*filters, in_channels_right=12*filters, out_channels_right=4*filters) self.cell_12 = FirstCell(in_channels_left=12*filters, out_channels_left=2*filters, in_channels_right=16*filters, out_channels_right=4*filters) self.cell_13 = NormalCell(in_channels_left=16*filters, out_channels_left=4*filters, in_channels_right=24*filters, out_channels_right=4*filters) self.cell_14 = NormalCell(in_channels_left=24*filters, out_channels_left=4*filters, in_channels_right=24*filters, out_channels_right=4*filters) self.cell_15 = NormalCell(in_channels_left=24*filters, out_channels_left=4*filters, in_channels_right=24*filters, out_channels_right=4*filters) self.relu = nn.ReLU() self.avg_pool = nn.AvgPool2d(7, stride=1, padding=0) self.dropout = nn.Dropout() self.last_linear = nn.Linear(24*filters, self.num_classes) def features(self, input): x_conv0 = self.conv0(input) x_stem_0 = self.cell_stem_0(x_conv0) x_stem_1 = self.cell_stem_1(x_conv0, x_stem_0) x_cell_0 = self.cell_0(x_stem_1, x_stem_0) x_cell_1 = self.cell_1(x_cell_0, x_stem_1) x_cell_2 = self.cell_2(x_cell_1, x_cell_0) x_cell_3 = self.cell_3(x_cell_2, x_cell_1) x_reduction_cell_0 = self.reduction_cell_0(x_cell_3, x_cell_2) x_cell_6 = self.cell_6(x_reduction_cell_0, x_cell_3) x_cell_7 = self.cell_7(x_cell_6, x_reduction_cell_0) x_cell_8 = self.cell_8(x_cell_7, x_cell_6) x_cell_9 = self.cell_9(x_cell_8, x_cell_7) x_reduction_cell_1 = self.reduction_cell_1(x_cell_9, x_cell_8) x_cell_12 = self.cell_12(x_reduction_cell_1, x_cell_9) x_cell_13 = self.cell_13(x_cell_12, x_reduction_cell_1) x_cell_14 = self.cell_14(x_cell_13, x_cell_12) x_cell_15 = self.cell_15(x_cell_14, x_cell_13) return x_cell_15 def logits(self, features): x = self.relu(features) x = self.avg_pool(x) x = x.view(x.size(0), -1) x = self.dropout(x) x = self.last_linear(x) return x def forward(self, input): x = self.features(input) x = self.logits(x) return x net5 = NASNetAMobile() net5.last_linear = nn.Linear(in_features=1056, out_features=1, bias=True) net5.load_state_dict(torch.load(CHECKPOINT5)) net5 = net5.to(device) net5.cuda() net5.eval() net9 = NASNetAMobile() net9.last_linear = nn.Linear(in_features=1056, out_features=1, bias=True) net9.load_state_dict(torch.load(CHECKPOINT9)) net9 = net9.to(device) net9.cuda() net9.eval()
Deepfake Detection Challenge
8,522,656
sub_2 = smpl_sub.copy() sub_2[CLASSES] = test_preds_arr.mean(axis=0 )<feature_engineering>
class CFG: seq_len=10 lstm_in = 16 lstm_out = 16 class LSTM_Model(nn.Module): def __init__(self): super(LSTM_Model, self ).__init__() self.cnn_net = mobilenetv3(mode='small', pretrained=False) self.cnn_net.classifier[1] = nn.Linear(in_features=1280, out_features=1) self.cnn_net.classifier[1] = nn.Linear(in_features=1280, out_features=CFG.lstm_in) self.lstm = nn.LSTM(CFG.lstm_in, CFG.lstm_out, bidirectional=False, batch_first=True) self.reg_layer = nn.Sequential(nn.Dropout(0.5), nn.Linear(CFG.lstm_out, 1)) def forward(self, x): n_samples = x.shape[0] x = x.view(-1, 3, 224, 224) x = self.cnn_net(x) x = x.view(n_samples, CFG.seq_len, -1) x,(h_n, h_c)= self.lstm(x) x = x[:,-1] y = self.reg_layer(x) return y net7 = LSTM_Model() net7.load_state_dict(torch.load(CHECKPOINT7)) net7 = net7.to(device) net7.cuda() net7.eval() net8 = LSTM_Model() net8.load_state_dict(torch.load(CHECKPOINT8)) net8 = net8.to(device) net8.cuda() net8.eval()
Deepfake Detection Challenge
8,522,656
sub[CLASSES] = 0.6 * sub[CLASSES] + 0.4 * sub_2[CLASSES]<save_to_csv>
Deepfake Detection Challenge
8,522,656
sub.to_csv("submission.csv", index=False )<import_modules>
net11 = torchvision.models.resnet18(pretrained=False) net11.fc = nn.Linear(in_features=512, out_features=1, bias=True) net11.load_state_dict(torch.load(CHECKPOINT11)) net11 = net11.to(device) net11.cuda() net11.eval()
Deepfake Detection Challenge
8,522,656
import pandas as pd from IPython.display import display from sklearn.feature_selection import VarianceThreshold from sklearn.ensemble import ExtraTreesClassifier from sklearn import svm from sklearn import preprocessing from sklearn.model_selection import train_test_split from sklearn.metrics import roc_auc_score, mean_squared_error import numpy as np<categorify>
import albumentations from albumentations.augmentations.transforms import ShiftScaleRotate, HorizontalFlip, RandomBrightnessContrast, MotionBlur, Blur, GaussNoise, JpegCompression
Deepfake Detection Challenge
8,522,656
types = {'ID':np.uint32, 'target':np.uint8, 'VAR_0002':np.uint16, 'VAR_0003':np.uint16, 'VAR_0532':np.uint8, 'VAR_0533':np.uint8, 'VAR_0534':np.uint8, 'VAR_0535':np.uint8, 'VAR_0536':np.uint8, 'VAR_0537':np.uint8,'VAR_0538':np.uint8, 'VAR_0539':np.uint8, 'VAR_0540':np.uint8, 'VAR_0545':np.uint16, 'VAR_0546':np.uint16, 'VAR_0547':np.uint16, 'VAR_0548':np.uint16, 'VAR_0549':np.uint16, 'VAR_0550':np.uint16, 'VAR_0551':np.uint16, 'VAR_0552':np.uint8, 'VAR_0553':np.uint8, 'VAR_0554':np.uint16, 'VAR_0555':np.uint16, 'VAR_0556':np.uint16, 'VAR_0557':np.uint16, 'VAR_0558':np.uint16, 'VAR_0559':np.uint8, 'VAR_0560':np.uint8, 'VAR_0561':np.uint16, 'VAR_0562':np.uint8, 'VAR_0563':np.uint8, 'VAR_0564':np.uint8, 'VAR_0565':np.uint8, 'VAR_0566':np.uint8, 'VAR_0567':np.uint8, 'VAR_0568':np.uint8, 'VAR_0569':np.uint8, 'VAR_0570':np.uint16, 'VAR_0572':np.uint8, 'VAR_0580':np.uint8, 'VAR_0581':np.uint8, 'VAR_0582':np.uint8, 'VAR_0604':np.uint8, 'VAR_0605':np.uint8, 'VAR_0606':np.uint8, 'VAR_0617':np.uint8, 'VAR_0618':np.uint8, 'VAR_0619':np.uint8, 'VAR_0620':np.uint8, 'VAR_0621':np.uint8, 'VAR_0622':np.uint8, 'VAR_0623':np.uint8, 'VAR_0624':np.uint8, 'VAR_0625':np.uint8, 'VAR_0626':np.uint8, 'VAR_0627':np.uint8, 'VAR_0637':np.uint8, 'VAR_0638':np.uint8, 'VAR_0639':np.uint8, 'VAR_0640':np.uint8, 'VAR_0646':np.uint8, 'VAR_0647':np.uint8, 'VAR_0657':np.uint8, 'VAR_0658':np.uint8, 'VAR_0662':np.uint8, 'VAR_0663':np.uint8, 'VAR_0664':np.uint8, 'VAR_0665':np.uint8, 'VAR_0666':np.uint8,'VAR_0667':np.uint8, 'VAR_0668':np.uint8, 'VAR_0685':np.uint8, 'VAR_0686':np.uint8, 'VAR_0689':np.uint8, 'VAR_0690':np.uint8, 'VAR_0696':np.uint8, 'VAR_0697':np.uint8, 'VAR_0703':np.uint8, 'VAR_0708':np.uint8, 'VAR_0709':np.uint8, 'VAR_0710':np.uint8, 'VAR_0711':np.uint8, 'VAR_0712':np.uint8, 'VAR_0713':np.uint8, 'VAR_0714':np.uint8, 'VAR_0715':np.uint8, 'VAR_0716':np.uint8, 'VAR_0717':np.uint8, 'VAR_0718':np.uint8, 'VAR_0719':np.uint8, 'VAR_0720':np.uint8, 'VAR_0721':np.uint8, 'VAR_0722':np.uint8, 'VAR_0723':np.uint8, 'VAR_0724':np.uint8, 'VAR_0725':np.uint8, 'VAR_0726':np.uint8, 'VAR_0727':np.uint8, 'VAR_0728':np.uint8, 'VAR_0729':np.uint8, 'VAR_0730':np.uint8, 'VAR_0731':np.uint8, 'VAR_0732':np.uint8, 'VAR_0733':np.uint8, 'VAR_0734':np.uint8, 'VAR_0735':np.uint8, 'VAR_0736':np.uint8, 'VAR_0737':np.uint8, 'VAR_0738':np.uint8, 'VAR_0739':np.uint8, 'VAR_0740':np.uint8, 'VAR_0741':np.uint8, 'VAR_0742':np.uint8, 'VAR_0743':np.uint8, 'VAR_0744':np.uint8, 'VAR_0745':np.uint8, 'VAR_0746':np.uint8, 'VAR_0747':np.uint8, 'VAR_0748':np.uint8, 'VAR_0749':np.uint8, 'VAR_0750':np.uint8, 'VAR_0751':np.uint8, 'VAR_0752':np.uint8, 'VAR_0753':np.uint8, 'VAR_0754':np.uint8, 'VAR_0755':np.uint8, 'VAR_0756':np.uint8, 'VAR_0758':np.uint8, 'VAR_0759':np.uint8, 'VAR_0760':np.uint8, 'VAR_0761':np.uint8, 'VAR_0762':np.uint8, 'VAR_0763':np.uint8, 'VAR_0764':np.uint8, 'VAR_0765':np.uint8, 'VAR_0766':np.uint8, 'VAR_0767':np.uint8, 'VAR_0768':np.uint8, 'VAR_0769':np.uint8, 'VAR_0770':np.uint8, 'VAR_0771':np.uint8, 'VAR_0772':np.uint8, 'VAR_0773':np.uint8, 'VAR_0774':np.uint8, 'VAR_0775':np.uint8, 'VAR_0776':np.uint8, 'VAR_0777':np.uint8, 'VAR_0778':np.uint8, 'VAR_0779':np.uint8, 'VAR_0780':np.uint8, 'VAR_0781':np.uint8, 'VAR_0782':np.uint8, 'VAR_0783':np.uint8, 'VAR_0784':np.uint8, 'VAR_0785':np.uint8, 'VAR_0786':np.uint8, 'VAR_0787':np.uint8, 'VAR_0788':np.uint8, 'VAR_0789':np.uint8, 'VAR_0790':np.uint8, 'VAR_0791':np.uint8, 'VAR_0792':np.uint8, 'VAR_0793':np.uint8, 'VAR_0794':np.uint8, 'VAR_0795':np.uint8, 'VAR_0796':np.uint8, 'VAR_0797':np.uint8, 'VAR_0798':np.uint8, 'VAR_0799':np.uint8, 'VAR_0800':np.uint8, 'VAR_0801':np.uint8, 'VAR_0802':np.uint8, 'VAR_0803':np.uint8, 'VAR_0804':np.uint8, 'VAR_0805':np.uint8, 'VAR_0806':np.uint8, 'VAR_0807':np.uint8, 'VAR_0808':np.uint8, 'VAR_0809':np.uint8, 'VAR_0810':np.uint8, 'VAR_0812':np.uint8, 'VAR_0813':np.uint8, 'VAR_0814':np.uint8, 'VAR_0815':np.uint8, 'VAR_0816':np.uint8, 'VAR_0817':np.uint8, 'VAR_0818':np.uint8, 'VAR_0819':np.uint8, 'VAR_0820':np.uint8, 'VAR_0821':np.uint8, 'VAR_0822':np.uint8, 'VAR_0823':np.uint8, 'VAR_0824':np.uint8, 'VAR_0825':np.uint8, 'VAR_0826':np.uint8, 'VAR_0827':np.uint8, 'VAR_0828':np.uint8, 'VAR_0829':np.uint8, 'VAR_0830':np.uint8, 'VAR_0831':np.uint8, 'VAR_0832':np.uint8, 'VAR_0833':np.uint8, 'VAR_0834':np.uint8, 'VAR_0835':np.uint8, 'VAR_0836':np.uint8, 'VAR_0837':np.uint8, 'VAR_0838':np.uint8, 'VAR_0839':np.uint8, 'VAR_0841':np.uint8, 'VAR_0842':np.uint8, 'VAR_0843':np.uint8, 'VAR_0844':np.uint8, 'VAR_0845':np.uint8, 'VAR_0846':np.uint8, 'VAR_0847':np.uint8, 'VAR_0848':np.uint8, 'VAR_0849':np.uint8, 'VAR_0850':np.uint8, 'VAR_0851':np.uint8, 'VAR_0852':np.uint8, 'VAR_0853':np.uint8, 'VAR_0854':np.uint8, 'VAR_0855':np.uint8, 'VAR_0856':np.uint8, 'VAR_0857':np.uint8, 'VAR_0859':np.uint8, 'VAR_0877':np.uint8, 'VAR_0878':np.uint8, 'VAR_0879':np.uint8, 'VAR_0885':np.uint8, 'VAR_0886':np.uint8, 'VAR_0911':np.uint8, 'VAR_0914':np.uint8, 'VAR_0915':np.uint8, 'VAR_0916':np.uint8, 'VAR_0923':np.uint8, 'VAR_0924':np.uint8, 'VAR_0925':np.uint8, 'VAR_0926':np.uint8, 'VAR_0927':np.uint8, 'VAR_0940':np.uint8, 'VAR_0945':np.uint8, 'VAR_0947':np.uint8, 'VAR_0952':np.uint8, 'VAR_0954':np.uint8, 'VAR_0959':np.uint8, 'VAR_0962':np.uint8, 'VAR_0963':np.uint8, 'VAR_0969':np.uint8, 'VAR_0973':np.uint8, 'VAR_0974':np.uint8, 'VAR_0975':np.uint8, 'VAR_0983':np.uint8, 'VAR_0984':np.uint8, 'VAR_0985':np.uint8, 'VAR_0986':np.uint8, 'VAR_0987':np.uint8, 'VAR_0988':np.uint8, 'VAR_0989':np.uint8, 'VAR_0990':np.uint8, 'VAR_0991':np.uint8, 'VAR_0992':np.uint8, 'VAR_0993':np.uint8, 'VAR_0994':np.uint8, 'VAR_0995':np.uint8, 'VAR_0996':np.uint8, 'VAR_0997':np.uint8, 'VAR_0998':np.uint8, 'VAR_0999':np.uint8, 'VAR_1000':np.uint8, 'VAR_1001':np.uint8, 'VAR_1002':np.uint8, 'VAR_1003':np.uint8, 'VAR_1004':np.uint8, 'VAR_1005':np.uint8, 'VAR_1006':np.uint8, 'VAR_1007':np.uint8, 'VAR_1008':np.uint8, 'VAR_1009':np.uint8, 'VAR_1010':np.uint8, 'VAR_1011':np.uint8, 'VAR_1012':np.uint8, 'VAR_1013':np.uint8, 'VAR_1014':np.uint8, 'VAR_1015':np.uint8, 'VAR_1016':np.uint8, 'VAR_1017':np.uint8, 'VAR_1018':np.uint8, 'VAR_1019':np.uint8, 'VAR_1020':np.uint8, 'VAR_1021':np.uint8, 'VAR_1022':np.uint8, 'VAR_1023':np.uint8, 'VAR_1024':np.uint8, 'VAR_1025':np.uint8, 'VAR_1026':np.uint8, 'VAR_1027':np.uint8, 'VAR_1028':np.uint8, 'VAR_1029':np.uint8, 'VAR_1030':np.uint8, 'VAR_1031':np.uint8, 'VAR_1032':np.uint8, 'VAR_1033':np.uint8, 'VAR_1034':np.uint8, 'VAR_1035':np.uint8, 'VAR_1036':np.uint8, 'VAR_1037':np.uint8, 'VAR_1038':np.uint8, 'VAR_1039':np.uint8, 'VAR_1040':np.uint8, 'VAR_1041':np.uint8, 'VAR_1042':np.uint8, 'VAR_1043':np.uint8, 'VAR_1044':np.uint8, 'VAR_1045':np.uint8, 'VAR_1046':np.uint8, 'VAR_1047':np.uint8, 'VAR_1048':np.uint8, 'VAR_1049':np.uint8, 'VAR_1050':np.uint8, 'VAR_1051':np.uint8, 'VAR_1052':np.uint8, 'VAR_1053':np.uint8, 'VAR_1054':np.uint8, 'VAR_1055':np.uint8, 'VAR_1056':np.uint8, 'VAR_1057':np.uint8, 'VAR_1058':np.uint8, 'VAR_1059':np.uint8, 'VAR_1060':np.uint8, 'VAR_1061':np.uint8, 'VAR_1062':np.uint8, 'VAR_1063':np.uint8, 'VAR_1064':np.uint8, 'VAR_1065':np.uint8, 'VAR_1066':np.uint8, 'VAR_1067':np.uint8, 'VAR_1068':np.uint8, 'VAR_1069':np.uint8, 'VAR_1070':np.uint8, 'VAR_1071':np.uint8, 'VAR_1072':np.uint8, 'VAR_1073':np.uint8, 'VAR_1080':np.uint8, 'VAR_1108':np.uint8, 'VAR_1109':np.uint8, 'VAR_1161':np.uint8, 'VAR_1162':np.uint8, 'VAR_1163':np.uint8, 'VAR_1164':np.uint8, 'VAR_1165':np.uint8, 'VAR_1166':np.uint8, 'VAR_1167':np.uint8, 'VAR_1168':np.uint8, 'VAR_1175':np.uint8, 'VAR_1176':np.uint8, 'VAR_1177':np.uint8, 'VAR_1178':np.uint8, 'VAR_1185':np.uint8, 'VAR_1186':np.uint8, 'VAR_1187':np.uint8, 'VAR_1188':np.uint8, 'VAR_1189':np.uint8, 'VAR_1190':np.uint8, 'VAR_1191':np.uint8, 'VAR_1192':np.uint8, 'VAR_1193':np.uint8, 'VAR_1194':np.uint8, 'VAR_1195':np.uint8, 'VAR_1196':np.uint8, 'VAR_1197':np.uint8, 'VAR_1198':np.uint8, 'VAR_1212':np.uint8, 'VAR_1213':np.uint8, 'VAR_1217':np.uint8, 'VAR_1218':np.uint8, 'VAR_1224':np.uint8, 'VAR_1225':np.uint8, 'VAR_1226':np.uint8, 'VAR_1229':np.uint8, 'VAR_1230':np.uint8, 'VAR_1231':np.uint8, 'VAR_1232':np.uint8, 'VAR_1233':np.uint8, 'VAR_1234':np.uint8, 'VAR_1235':np.uint8, 'VAR_1236':np.uint8, 'VAR_1237':np.uint8, 'VAR_1238':np.uint8, 'VAR_1239':np.uint8, 'VAR_1267':np.uint8, 'VAR_1268':np.uint8, 'VAR_1269':np.uint8, 'VAR_1270':np.uint8, 'VAR_1271':np.uint8, 'VAR_1272':np.uint8, 'VAR_1273':np.uint8, 'VAR_1274':np.uint8, 'VAR_1275':np.uint8, 'VAR_1276':np.uint8, 'VAR_1277':np.uint8, 'VAR_1278':np.uint8, 'VAR_1279':np.uint8, 'VAR_1280':np.uint8, 'VAR_1281':np.uint8, 'VAR_1282':np.uint8, 'VAR_1283':np.uint8, 'VAR_1284':np.uint8, 'VAR_1285':np.uint8, 'VAR_1286':np.uint8, 'VAR_1287':np.uint8, 'VAR_1288':np.uint8, 'VAR_1289':np.uint8, 'VAR_1290':np.uint8, 'VAR_1291':np.uint8, 'VAR_1292':np.uint8, 'VAR_1293':np.uint8, 'VAR_1294':np.uint8, 'VAR_1295':np.uint8, 'VAR_1296':np.uint8, 'VAR_1297':np.uint8, 'VAR_1298':np.uint8, 'VAR_1299':np.uint8, 'VAR_1300':np.uint8, 'VAR_1301':np.uint8, 'VAR_1302':np.uint8, 'VAR_1303':np.uint8, 'VAR_1304':np.uint8, 'VAR_1305':np.uint8, 'VAR_1306':np.uint8, 'VAR_1307':np.uint8, 'VAR_1338':np.uint8, 'VAR_1339':np.uint8, 'VAR_1340':np.uint8, 'VAR_1345':np.uint8, 'VAR_1346':np.uint8, 'VAR_1347':np.uint8, 'VAR_1348':np.uint8, 'VAR_1349':np.uint8, 'VAR_1350':np.uint8, 'VAR_1351':np.uint8, 'VAR_1352':np.uint8, 'VAR_1359':np.uint8, 'VAR_1360':np.uint8, 'VAR_1361':np.uint8, 'VAR_1362':np.uint8, 'VAR_1363':np.uint8, 'VAR_1364':np.uint8, 'VAR_1365':np.uint8, 'VAR_1366':np.uint8, 'VAR_1367':np.uint8, 'VAR_1368':np.uint8, 'VAR_1369':np.uint8, 'VAR_1386':np.uint8, 'VAR_1387':np.uint8, 'VAR_1388':np.uint8, 'VAR_1389':np.uint8, 'VAR_1392':np.uint8, 'VAR_1393':np.uint8, 'VAR_1394':np.uint8, 'VAR_1395':np.uint8, 'VAR_1396':np.uint8, 'VAR_1404':np.uint8, 'VAR_1405':np.uint8, 'VAR_1406':np.uint8, 'VAR_1407':np.uint8, 'VAR_1408':np.uint8, 'VAR_1409':np.uint8, 'VAR_1410':np.uint8, 'VAR_1411':np.uint8, 'VAR_1412':np.uint8, 'VAR_1413':np.uint8, 'VAR_1414':np.uint8, 'VAR_1415':np.uint8, 'VAR_1416':np.uint8, 'VAR_1417':np.uint8, 'VAR_1427':np.uint8, 'VAR_1428':np.uint8, 'VAR_1429':np.uint8, 'VAR_1430':np.uint8, 'VAR_1431':np.uint8, 'VAR_1432':np.uint8, 'VAR_1433':np.uint8, 'VAR_1434':np.uint8, 'VAR_1435':np.uint8, 'VAR_1449':np.uint8, 'VAR_1450':np.uint8, 'VAR_1456':np.uint8, 'VAR_1457':np.uint8, 'VAR_1458':np.uint8, 'VAR_1459':np.uint8, 'VAR_1460':np.uint8, 'VAR_1461':np.uint8, 'VAR_1462':np.uint8, 'VAR_1463':np.uint8, 'VAR_1464':np.uint8, 'VAR_1465':np.uint8, 'VAR_1466':np.uint8, 'VAR_1467':np.uint8, 'VAR_1468':np.uint8, 'VAR_1469':np.uint8, 'VAR_1470':np.uint8, 'VAR_1471':np.uint8, 'VAR_1472':np.uint8, 'VAR_1473':np.uint8, 'VAR_1474':np.uint8, 'VAR_1475':np.uint8, 'VAR_1476':np.uint8, 'VAR_1477':np.uint8, 'VAR_1478':np.uint8, 'VAR_1479':np.uint8, 'VAR_1480':np.uint8, 'VAR_1481':np.uint8, 'VAR_1482':np.uint8, 'VAR_1532':np.uint8, 'VAR_1533':np.uint8, 'VAR_1534':np.uint8, 'VAR_1535':np.uint8, 'VAR_1537':np.uint8, 'VAR_1538':np.uint8, 'VAR_1539':np.uint8, 'VAR_1540':np.uint8, 'VAR_1542':np.uint8, 'VAR_1543':np.uint8, 'VAR_1544':np.uint8, 'VAR_1545':np.uint8, 'VAR_1546':np.uint8, 'VAR_1547':np.uint8, 'VAR_1548':np.uint8, 'VAR_1549':np.uint8, 'VAR_1551':np.uint8, 'VAR_1552':np.uint8, 'VAR_1553':np.uint8, 'VAR_1554':np.uint8, 'VAR_1556':np.uint8, 'VAR_1557':np.uint8, 'VAR_1558':np.uint8, 'VAR_1559':np.uint8, 'VAR_1561':np.uint8, 'VAR_1562':np.uint8, 'VAR_1563':np.uint8, 'VAR_1564':np.uint8, 'VAR_1565':np.uint8, 'VAR_1566':np.uint8, 'VAR_1567':np.uint8, 'VAR_1568':np.uint8, 'VAR_1569':np.uint8, 'VAR_1570':np.uint8, 'VAR_1571':np.uint8, 'VAR_1572':np.uint8, 'VAR_1574':np.uint8, 'VAR_1575':np.uint8, 'VAR_1576':np.uint8, 'VAR_1577':np.uint8, 'VAR_1578':np.uint8, 'VAR_1579':np.uint8, 'VAR_1583':np.uint8, 'VAR_1584':np.uint8, 'VAR_1585':np.uint8, 'VAR_1586':np.uint8, 'VAR_1587':np.uint8, 'VAR_1588':np.uint8, 'VAR_1589':np.uint8, 'VAR_1590':np.uint8, 'VAR_1591':np.uint8, 'VAR_1592':np.uint8, 'VAR_1593':np.uint8, 'VAR_1594':np.uint8, 'VAR_1595':np.uint8, 'VAR_1596':np.uint8, 'VAR_1597':np.uint8, 'VAR_1598':np.uint8, 'VAR_1599':np.uint8, 'VAR_1600':np.uint8, 'VAR_1601':np.uint8, 'VAR_1602':np.uint8, 'VAR_1603':np.uint8, 'VAR_1604':np.uint8, 'VAR_1605':np.uint8, 'VAR_1606':np.uint8, 'VAR_1607':np.uint8, 'VAR_1608':np.uint8, 'VAR_1609':np.uint8, 'VAR_1610':np.uint8, 'VAR_1656':np.uint8, 'VAR_1657':np.uint8, 'VAR_1658':np.uint8, 'VAR_1659':np.uint8, 'VAR_1660':np.uint8, 'VAR_1661':np.uint8, 'VAR_1662':np.uint8, 'VAR_1663':np.uint8, 'VAR_1664':np.uint8, 'VAR_1665':np.uint8, 'VAR_1666':np.uint8, 'VAR_1667':np.uint8, 'VAR_1668':np.uint8, 'VAR_1669':np.uint8, 'VAR_1670':np.uint8, 'VAR_1671':np.uint8, 'VAR_1672':np.uint8, 'VAR_1673':np.uint8, 'VAR_1674':np.uint8, 'VAR_1675':np.uint8, 'VAR_1676':np.uint8, 'VAR_1677':np.uint8, 'VAR_1678':np.uint8, 'VAR_1679':np.uint8, 'VAR_1680':np.uint8, 'VAR_1681':np.uint8, 'VAR_1682':np.uint8, 'VAR_1683':np.uint8, 'VAR_1713':np.uint8, 'VAR_1714':np.uint8, 'VAR_1721':np.uint8, 'VAR_1722':np.uint8, 'VAR_1723':np.uint8, 'VAR_1724':np.uint8, 'VAR_1725':np.uint8, 'VAR_1726':np.uint8, 'VAR_1727':np.uint8, 'VAR_1728':np.uint8, 'VAR_1740':np.uint8, 'VAR_1741':np.uint8, 'VAR_1742':np.uint8, 'VAR_1743':np.uint8, 'VAR_1744':np.uint8, 'VAR_1745':np.uint8, 'VAR_1746':np.uint8, 'VAR_1752':np.uint8, 'VAR_1753':np.uint8, 'VAR_1760':np.uint8, 'VAR_1761':np.uint8, 'VAR_1762':np.uint8, 'VAR_1763':np.uint8, 'VAR_1764':np.uint8, 'VAR_1765':np.uint8, 'VAR_1766':np.uint8, 'VAR_1767':np.uint8, 'VAR_1768':np.uint8, 'VAR_1769':np.uint8, 'VAR_1770':np.uint8, 'VAR_1771':np.uint8, 'VAR_1772':np.uint8, 'VAR_1773':np.uint8, 'VAR_1774':np.uint8, 'VAR_1775':np.uint8, 'VAR_1776':np.uint8, 'VAR_1777':np.uint8, 'VAR_1778':np.uint8, 'VAR_1779':np.uint8, 'VAR_1780':np.uint8, 'VAR_1781':np.uint8, 'VAR_1782':np.uint8, 'VAR_1783':np.uint8, 'VAR_1784':np.uint8, 'VAR_1785':np.uint8, 'VAR_1786':np.uint8, 'VAR_1787':np.uint8, 'VAR_1788':np.uint8, 'VAR_1789':np.uint8, 'VAR_1790':np.uint8, 'VAR_1791':np.uint8, 'VAR_1792':np.uint8, 'VAR_1793':np.uint8, 'VAR_1794':np.uint8, 'VAR_1843':np.uint8, 'VAR_1844':np.uint8, 'VAR_1853':np.uint8, 'VAR_1854':np.uint8, 'VAR_1855':np.uint8, 'VAR_1856':np.uint8, 'VAR_1857':np.uint8, 'VAR_1866':np.uint8, 'VAR_1867':np.uint8, 'VAR_1872':np.uint8, 'VAR_1873':np.uint8, 'VAR_1874':np.uint8, 'VAR_1875':np.uint8, 'VAR_1876':np.uint8, 'VAR_1877':np.uint8, 'VAR_1878':np.uint8, 'VAR_1879':np.uint8, 'VAR_1880':np.uint8, 'VAR_1881':np.uint8, 'VAR_1882':np.uint8, 'VAR_1883':np.uint8, 'VAR_1884':np.uint8, 'VAR_1885':np.uint8, 'VAR_1886':np.uint8, 'VAR_1887':np.uint8, 'VAR_1888':np.uint8, 'VAR_1903':np.uint8, 'VAR_1904':np.uint8, 'VAR_1905':np.uint8, 'VAR_1906':np.uint8, 'VAR_1907':np.uint8, 'VAR_1908':np.uint8, 'VAR_1909':np.uint8, 'VAR_1910':np.uint8, 'VAR_1920':np.uint8, 'VAR_1921':np.uint8, 'VAR_1925':np.uint8, 'VAR_1926':np.uint8, 'VAR_1927':np.uint8, 'VAR_1928':np.uint16, 'VAR_1930':np.uint16}<load_from_csv>
def predict_on_video(model, model2, model3, model4, model5, model6, model7, model8, model9, model10, model11, video_path): try: x, x_sqr, x_299_sqr, x_lstm, frame_skip = extract_frames(video_path) if x is None or x_sqr is None: return 0.5 else: with torch.no_grad() : y_pred = model(x.to(device)) y_pred = torch.sigmoid(y_pred.squeeze().mean()) y_pred2 = model2(x_sqr.to(device)) y_pred2 = torch.sigmoid(y_pred2.squeeze().mean()) y_pred3 = model3(x_sqr.to(device)) y_pred3 = torch.sigmoid(y_pred3.squeeze().mean()) y_pred4 = model4(x_299_sqr.to(device)) y_pred4 = torch.sigmoid(y_pred4.squeeze().mean()) y_pred5 = model5(x.to(device)) y_pred5 = torch.sigmoid(y_pred5.squeeze().mean()) y_pred6 = model6(x_sqr.to(device)) y_pred6 = torch.sigmoid(y_pred6.squeeze().mean()) seq_len = len(x_lstm)//2 x_sqr_1, x_sqr_2 = x_lstm[:seq_len], x_lstm[seq_len:] skip = max(1, seq_len//15) start_indices = list(range(0, seq_len, skip)) gap = max(1, int(np.round(20/frame_skip))) for ss, start in enumerate(start_indices): indices = list(range(start, 500, gap)) indices = [i%seq_len for i in indices] indices = indices[:CFG.seq_len] seqs1, seqs2 = x_sqr_1[indices].unsqueeze(0), x_sqr_2[indices].unsqueeze(0) if ss == 0: batch_seqs_1 = seqs1 batch_seqs_2 = seqs2 else: batch_seqs_1 = torch.cat(( batch_seqs_1, seqs1), 0) batch_seqs_2 = torch.cat(( batch_seqs_2, seqs2), 0) batch_seqs = torch.cat(( batch_seqs_1, batch_seqs_2), 0) y_pred7 = model7(batch_seqs.to(device)) y_pred7 = torch.sigmoid(y_pred7.squeeze().mean()) y_pred8 = model8(batch_seqs.to(device)) y_pred8 = torch.sigmoid(y_pred8.squeeze().mean()) y_pred9 = model9(x_sqr.to(device)) y_pred9 = torch.sigmoid(y_pred9.squeeze().mean()) y_pred10 = model10(x_299_sqr.to(device)) y_pred10 = torch.sigmoid(y_pred10.squeeze().mean()) y_pred11 = model11(x.to(device)) y_pred11 = torch.sigmoid(y_pred11.squeeze().mean()) w = [3, 2, 8, 2, 2, 2, 4, 4, 2, 2, 3] return(w[0]*y_pred.item() + w[1]*y_pred2.item() + w[2]*y_pred3.item() + w[3]*y_pred4.item() + w[4]*y_pred5.item() + w[5]*y_pred6.item() + w[6]*y_pred7.item() + w[7]*y_pred8.item() + w[8]*y_pred9.item() + w[9]*y_pred10.item() + w[10]*y_pred11.item())/sum(w) except Exception as e: print("Prediction error on video", e) return 0.5 return 0.5
Deepfake Detection Challenge
8,522,656
train = pd.read_csv(".. /input/springleaf-marketing-response/train.csv.zip") mixCol = [8,9,10,11,12,18,19,20,21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 38, 39, 40, 41, 42, 43, 44, 45, 73, 74, 98, 99, 100, 106, 107, 108, 156, 157, 158, 159, 166, 167, 168, 169, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 202, 205, 206, 207, 208, 209, 210, 211, 212, 213, 214, 215, 216, 218, 219, 220, 221, 222, 223, 224, 225, 240, 371, 372, 373, 374, 375, 376, 377, 378, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 399, 400, 401, 402, 403, 404, 405, 406, 427, 428, 429, 430, 431, 432, 433, 434, 435, 436, 437, 438, 439, 440, 441, 442, 443, 444, 445, 446, 447, 448, 449, 450, 451, 452, 453, 454, 455, 456, 457, 458, 459, 460, 461, 462, 463, 464, 465, 466, 467, 468, 469, 470, 471, 472, 473, 474, 475, 476, 477, 478, 479, 480, 481, 492, 493, 494, 495, 496, 497, 498, 499, 500, 501, 502, 503, 504, 505, 506, 507, 508, 509, 510, 511, 512, 513, 518, 519, 520, 521, 522, 523, 524, 525, 526, 527, 528, 529, 530, 840] alphaCol = [283, 305, 325, 352, 353, 354, 1934] placeCol = [200, 274, 342] dtCol = [75, 204, 217] selectColumns = [] rmCol = mixCol+alphaCol+placeCol+dtCol for i in range(1,1935): if i not in rmCol: selectColumns.append(i) cols = [str(n ).zfill(4)for n in selectColumns] strColName = ['VAR_' + strNum for strNum in cols]<load_from_csv>
class FastMTCNN(object): def __init__(self, resize=1, *args, **kwargs): self.resize = resize self.mtcnn = MTCNN(*args, **kwargs) def __call__(self, frames): if self.resize != 1: frames = [f.resize([int(d * self.resize)for d in f.size])for f in frames] boxes, probs = self.mtcnn.detect(frames) boxes = [b.astype(int ).tolist() if b is not None and type(b)==np.ndarray else [] for b in boxes] probs = [b.tolist() if b is not None and type(b)==np.ndarray else [] for b in probs] return boxes, probs fast_mtcnn = FastMTCNN( resize=SCALE, margin=14, keep_all=True, device=device )
Deepfake Detection Challenge
8,522,656
nrows = 500 trainData = pd.read_csv(".. /input/springleaf-marketing-response/train.csv.zip", skiprows=[107], usecols=strColName, nrows=nrows, dtype=types) label = pd.read_csv(".. /input/springleaf-marketing-response/train.csv.zip", skiprows=[107], usecols=['target'], nrows=nrows) testData = pd.read_csv(".. /input/springleaf-marketing-response/test.csv.zip", skiprows=[107], usecols=strColName, nrows=nrows) numericFeatures = trainData._get_numeric_data() removeNA = numericFeatures.fillna(0) sel = VarianceThreshold(threshold= (.8 *(1 -.8))) features = sel.fit_transform(removeNA) y = np.array(label ).ravel() X_scaled = preprocessing.scale(features) normalizer = preprocessing.Normalizer().fit(X_scaled) X_norm = normalizer.transform(X_scaled) X_train, X_test, y_train, y_test = train_test_split(X_norm, y, test_size=0.33, random_state=42) y_train = np.array(y_train) <train_model>
test_videos = sorted([x for x in os.listdir(TEST_DIR)if x[-4:] == ".mp4"]) len(test_videos )
Deepfake Detection Challenge
8,522,656
clf = svm.SVC(C=1.0, kernel='linear', degree=10, gamma=1.00, coef0=0.0, shrinking=True, probability=False, tol=0.001, cache_size=1000, class_weight=None, verbose=False, max_iter=-1, random_state=None) clf.fit(X_train, y_train) predictions = clf.predict(X_test) print('roc_auc_score', roc_auc_score(y_test, predictions)) print('RMSE', mean_squared_error(y_test, predictions))<train_on_grid>
def predict_on_video_set(model, model2, model3, model4, model5, model6, model7, model8, model9, model10, model11, videos, num_workers): def process_file(i): filename = videos[i] y_pred = predict_on_video(model, model2, model3, model4, model5, model6, model7, model8, model9, model10, model11, os.path.join(TEST_DIR, filename)) print(i, y_pred) return y_pred with ThreadPoolExecutor(max_workers=num_workers)as ex: predictions = ex.map(process_file, range(len(videos))) return list(predictions) predictions = predict_on_video_set(net, net2, net3, net4, net5, net6, net7, net8, net9, net10, net11, test_videos, num_workers=NUM_WORKERS )
Deepfake Detection Challenge
8,522,656
testData = pd.read_csv(".. /input/springleaf-marketing-response/test.csv.zip", usecols=strColName, engine='python', dtype=types) numericFeatures = testData._get_numeric_data() removeNA = numericFeatures.fillna(0) features = sel.transform(removeNA) y = np.array(label ).ravel() X_scaled = preprocessing.scale(features) normalizer = preprocessing.Normalizer().fit(X_scaled) X_norm = normalizer.transform(X_scaled) <predict_on_test>
predictions = np.clip(predictions, 0.005, 0.995) submission_df = pd.DataFrame({"filename": test_videos, "label": predictions}) submission_df.to_csv("submission.csv", index=False )
Deepfake Detection Challenge
8,681,162
predictions = clf.predict(X_norm )<load_from_csv>
import random import re from copy import deepcopy from typing import Union, List, Tuple, Optional, Callable from collections import OrderedDict, defaultdict import math import cv2 import torch import torch.nn as nn from torch.utils.data import Dataset,DataLoader from torch.utils.data.sampler import SequentialSampler, RandomSampler from torchvision import transforms, models from torchvision.transforms import Normalize from tqdm import tqdm from sklearn.cluster import DBSCAN
Deepfake Detection Challenge
8,681,162
df_submit = pd.read_csv('.. /input/springleaf-marketing-response/sample_submission.csv.zip') df_submit['target'] = predictions<save_to_csv>
TARGET_H, TARGET_W = 224, 224 FRAMES_PER_VIDEO = 30 TEST_VIDEOS_PATH = '.. /input/deepfake-detection-challenge/test_videos' NN_MODEL_PATHS = [ '.. /input/kdold-deepfake-effb2/fold0-effb2-000epoch.pt', '.. /input/kdold-deepfake-effb2/fold0-effb2-001epoch.pt', '.. /input/kdold-deepfake-effb2/fold0-effb2-002epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold0-flip-effb2-000epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold0-flip-effb2-001epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold0-flip-effb2-002epoch.pt', '.. /input/kdold-deepfake-effb2/fold1-effb2-000epoch.pt', '.. /input/kdold-deepfake-effb2/fold1-effb2-001epoch.pt', '.. /input/kdold-deepfake-effb2/fold1-effb2-002epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold1-flip-effb2-000epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold1-flip-effb2-001epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold1-flip-effb2-002epoch.pt', '.. /input/kdold-deepfake-effb2/fold2-effb2-000epoch.pt', '.. /input/kdold-deepfake-effb2/fold2-effb2-001epoch.pt', '.. /input/kdold-deepfake-effb2/fold2-effb2-002epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold2-flip-effb2-000epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold2-flip-effb2-001epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold2-flip-effb2-002epoch.pt', '.. /input/kdold-deepfake-effb2/fold3-effb2-000epoch.pt', '.. /input/kdold-deepfake-effb2/fold3-effb2-001epoch.pt', '.. /input/kdold-deepfake-effb2/fold3-effb2-002epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold3-flip-effb2-000epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold3-flip-effb2-001epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold3-flip-effb2-002epoch.pt', '.. /input/kdold-deepfake-effb2/fold4-effb2-000epoch.pt', '.. /input/kdold-deepfake-effb2/fold4-effb2-001epoch.pt', '.. /input/kdold-deepfake-effb2/fold4-effb2-002epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold4-flip-effb2-000epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold4-flip-effb2-001epoch.pt', '.. /input/kfolddeepfakeeffb2-flip/fold4-flip-effb2-002epoch.pt', ]
Deepfake Detection Challenge
8,681,162
df_submit.to_csv('submission.csv', index=False )<set_options>
SEED = 42 def seed_everything(seed): random.seed(seed) os.environ['PYTHONHASHSEED'] = str(seed) np.random.seed(seed) torch.manual_seed(seed) torch.cuda.manual_seed(seed) torch.backends.cudnn.deterministic = False torch.backends.cudnn.benchmark = True seed_everything(SEED )
Deepfake Detection Challenge
8,681,162
%matplotlib inline<import_modules>
!pip install.. /input/pytorchefficientnet/EfficientNet-PyTorch-master > /dev/null def get_net() : net = EfficientNet.from_name('efficientnet-b2') net._fc = nn.Linear(in_features=net._fc.in_features, out_features=2, bias=True) return net
Deepfake Detection Challenge
8,681,162
from xgboost import XGBRegressor from sklearn.kernel_ridge import KernelRidge from sklearn.model_selection import GridSearchCV from sklearn.feature_selection import RFECV from sklearn.preprocessing import StandardScaler from sklearn.metrics import mean_squared_error, make_scorer from mlxtend.preprocessing import minmax_scaling<set_options>
class DatasetRetriever(Dataset): def __init__(self, df): self.video_paths = df['video_path'] self.filenames = df.index self.face_dr = FaceDetector(frames_per_video=FRAMES_PER_VIDEO) mean = [0.485, 0.456, 0.406] std = [0.229, 0.224, 0.225] self.normalize_transform = Normalize(mean, std) self.video_reader = VideoReader() self.video_read_fn = lambda x: self.video_reader.read_frames(x, num_frames=FRAMES_PER_VIDEO) def __len__(self): return self.filenames.shape[0] def __getitem__(self, idx): video_path = self.video_paths[idx] filename = self.filenames[idx] my_frames, my_idxs = self.video_read_fn(video_path) faces = self.face_dr.get_faces( my_frames, my_idxs, 0.7, 0.7, 0.7, 0.6 ) n = len(faces) video = torch.zeros(( n, 3, TARGET_H, TARGET_W)) for i, face in enumerate(faces[:n]): face = 255 - face face = face.astype(np.float32)/255. face = torch.tensor(face) face = face.permute(2,0,1) face = self.normalize_transform(face) video[i] = face return filename, video
Deepfake Detection Challenge
8,681,162
warnings.filterwarnings("ignore" )<count_unique_values>
class DeepFakePredictor: def __init__(self): self.models = [self.prepare_model(get_net() , path)for path in NN_MODEL_PATHS] self.models_count = len(self.models) def predict(self, dataset): result = [] with torch.no_grad() : for filename, video in dataset: video = video.to(self.device, dtype=torch.float32) try: label = self.predict_ensemble(video) except Exception as e: print(f'Warning! {e}, {type(e)}') label = 0.5 result.append({ 'filename': filename, 'label': label, }) return pd.DataFrame(result ).set_index('filename') def prepare_model(self, model, path): self.device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') model.to(self.device); if torch.cuda.is_available() : model = model.cuda() if torch.cuda.is_available() : checkpoint = torch.load(path) else: checkpoint = torch.load(path, map_location=torch.device('cpu')) model.load_state_dict(checkpoint['model_state_dict']) model.eval() print(f'Model prepared.Device is {self.device}') return model @staticmethod def net_forward(net, inputs): bs = inputs.size(0) x = net.extract_features(inputs) x = net._avg_pooling(x) emb = x.view(bs, -1) x = net._dropout(emb) x = net._fc(x) return emb, x def postprocess(self, embs, predictions): clusters = defaultdict(list) for prediction, cluster_id in zip(predictions, DBSCAN(eps=1.2, min_samples=1 ).fit_predict(embs)) : clusters[cluster_id].append(prediction) sorted_clusters = sorted(clusters.items() , key=lambda x: -len(x[1])) if len(sorted_clusters)< 2: return sorted_clusters[0][1] if len(sorted_clusters[1][1])/ len(predictions)> 0.25: return sorted_clusters[0][1] + sorted_clusters[1][1] return sorted_clusters[0][1] def predict_ensemble(self, video): embs, predictions = 0, 0 for model in self.models: emb, prediction = self.net_forward(model, video) predictions += prediction / self.models_count embs += emb / self.models_count predictions = nn.functional.softmax(predictions, dim=1 ).data.cpu().numpy() [:,1] embs = embs.cpu().numpy() predictions = self.postprocess(embs, predictions) return np.mean(predictions )
Deepfake Detection Challenge
8,681,162
def show_uniqs(cols): for col in cols: print(col) show_uniq(col) print('=======================================' )<load_from_csv>
deep_fake_predictor = DeepFakePredictor()
Deepfake Detection Challenge
8,681,162
train = pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/train.csv' )<count_missing_values>
def process_dfs(df, num_workers=2): def process_df(sub_df): dataset = DatasetRetriever(sub_df) result = deep_fake_predictor.predict(dataset) return result with ThreadPoolExecutor(max_workers=num_workers)as ex: results = ex.map(process_df, np.split(df, num_workers)) return results
Deepfake Detection Challenge
8,681,162
<define_variables><EOS>
result.to_csv('submission.csv' )
Deepfake Detection Challenge
8,236,131
<SOS> metric: LogLoss Kaggle data source: deepfake-detection-challenge<load_from_csv>
!pip install.. /input/pytorchcv/pytorchcv-0.0.55-py2.py3-none-any.whl --quiet
Deepfake Detection Challenge
8,236,131
test = pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/test.csv' )<count_missing_values>
device = 'cuda' if torch.cuda.is_available() else 'cpu'
Deepfake Detection Challenge
8,236,131
count_empty_columns_test = test[empty_columns_test].isnull().sum(axis = 0) count_empty_columns_test<define_variables>
def gem(x, p=3, eps=1e-6): return F.avg_pool2d(x.clamp(min=eps ).pow(p),(x.size(-2), x.size(-1)) ).pow(1./p) class GeM(nn.Module): def __init__(self, p=3, eps=1e-6): super(GeM,self ).__init__() self.p = Parameter(torch.ones(1)*p) self.eps = eps def forward(self, x): return gem(x, p=self.p, eps=self.eps) def __repr__(self): return self.__class__.__name__ + '(' + 'p=' + '{:.4f}'.format(self.p.data.tolist() [0])+ ', ' + 'eps=' + str(self.eps)+ ')' class MishFunction(torch.autograd.Function): @staticmethod def forward(ctx, x): ctx.save_for_backward(x) return x * torch.tanh(F.softplus(x)) @staticmethod def backward(ctx, grad_output): x = ctx.saved_variables[0] sigmoid = torch.sigmoid(x) tanh_sp = torch.tanh(F.softplus(x)) return grad_output *(tanh_sp + x * sigmoid *(1 - tanh_sp * tanh_sp)) class Mish(nn.Module): def forward(self, x): return MishFunction.apply(x) def to_Mish(model): for child_name, child in model.named_children() : if isinstance(child, nn.ReLU): setattr(model, child_name, Mish()) else: to_Mish(child) class Head(torch.nn.Module): def __init__(self, in_f, out_f, hidden): super(Head, self ).__init__() self.f = nn.Flatten() self.l = nn.Linear(in_f, hidden) self.m = Mish() self.d = nn.Dropout(0.75) self.o = nn.Linear(hidden, out_f) self.b1 = nn.BatchNorm1d(in_f) self.b2 = nn.BatchNorm1d(hidden) self.r = nn.ReLU() def forward(self, x): x = self.f(x) x = self.b1(x) x = self.d(x) x = self.l(x) x = self.r(x) x = self.b2(x) x = self.d(x) out = self.o(x) return out class FCN(torch.nn.Module): def __init__(self, base, in_f, hidden): super(FCN, self ).__init__() self.base = base self.h1 = Head(in_f, 1, hidden) def forward(self, x): x = self.base(x) return self.h1(x) net = [] model = get_model("xception", pretrained=False) model = nn.Sequential(*list(model.children())[:-1]) model[0].final_block.pool = nn.Sequential(nn.AdaptiveAvgPool2d(1)) model = FCN(model, 2048, 512) model = model.cuda() model.load_state_dict(torch.load('.. /input/deepfake-models/model2.pth')) net.append(model) model = get_model("efficientnet_b1", pretrained=False) model = nn.Sequential(*list(model.children())[:-1]) model[0].final_pool = nn.Sequential(nn.AdaptiveAvgPool2d(1)) model = FCN(model, 1280, 512) model = model.cuda() model.load_state_dict(torch.load('.. /input/deepfake-models/model (.259 ).pth')) net.append(model) model = get_model("efficientnet_b1", pretrained=False) model = nn.Sequential(*list(model.children())[:-1]) model[0].final_pool = nn.Sequential(nn.AdaptiveAvgPool2d(1)) model = FCN(model, 1280, 512) model = model.cuda() model.load_state_dict(torch.load('.. /input/deepfake-models/model_16 (.2755 ).pth')) net.append(model)
Deepfake Detection Challenge
8,236,131
for col in count_empty_columns_test.index: if col not in count_empty_columns_train: print(col )<define_variables>
Deepfake Detection Challenge
8,236,131
garage_colums = [] regexp = re.compile(r"([-a-zA-Z]+)?"+r"Garage"+r"([-a-zA-Z]+)?") for col in train.columns: if regexp.search(col): garage_colums.append(col) garage_colums = np.array(garage_colums) garage_colums<data_type_conversions>
mean = [0.485, 0.456, 0.406] std = [0.229, 0.224, 0.225] normalize_transform = Normalize(mean, std )
Deepfake Detection Challenge
8,236,131
for col in garage_num_cols: train[col] = train[col].fillna(0) test[col] = test[col].fillna(0 )<count_unique_values>
detection_graph = tf.Graph() with detection_graph.as_default() : od_graph_def = tf.compat.v1.GraphDef() with tf.io.gfile.GFile('.. /input/mobilenet-face/frozen_inference_graph_face.pb', 'rb')as fid: serialized_graph = fid.read() od_graph_def.ParseFromString(serialized_graph) tf.import_graph_def(od_graph_def, name='') config = tf.compat.v1.ConfigProto() config.gpu_options.allow_growth = True sess=tf.compat.v1.Session(graph=detection_graph, config=config) image_tensor = detection_graph.get_tensor_by_name('image_tensor:0') boxes_tensor = detection_graph.get_tensor_by_name('detection_boxes:0') scores_tensor = detection_graph.get_tensor_by_name('detection_scores:0') num_detections = detection_graph.get_tensor_by_name('num_detections:0') def get_mobilenet_face(image): global boxes,scores,num_detections (im_height,im_width)=image.shape[:-1] imgs=np.array([image]) (boxes, scores)= sess.run( [boxes_tensor, scores_tensor], feed_dict={image_tensor: imgs}) max_=np.where(scores==scores.max())[0][0] box=boxes[0][max_] ymin, xmin, ymax, xmax = box (left, right, top, bottom)=(xmin * im_width, xmax * im_width, ymin * im_height, ymax * im_height) left, right, top, bottom = int(left), int(right), int(top), int(bottom) return(left, right, top, bottom) def crop_image(frame,bbox): left, right, top, bottom=bbox return frame[top:bottom,left:right] def get_img(frame): return cv2.resize(crop_image(frame,get_mobilenet_face(frame)) ,(160,160)) def detect_video(video): capture = cv2.VideoCapture(video) v_len = int(capture.get(cv2.CAP_PROP_FRAME_COUNT)) frame_idxs = np.linspace(0,v_len,frame_count, endpoint=False, dtype=np.int) imgs=[] i=0 for frame_idx in range(int(v_len)) : ret = capture.grab() if not ret: pass if frame_idx >= frame_idxs[i]: ret, frame = capture.retrieve() if not ret or frame is None: pass else: frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) try: face=get_img(frame) except Exception as err: print(err) continue face = cv2.cvtColor(face, cv2.COLOR_BGR2RGB) imgs.append(face) i += 1 if i >= len(frame_idxs): break if len(imgs)<frame_count: return None return imgs
Deepfake Detection Challenge
8,236,131
<define_variables><EOS>
probs = np.asarray(probs) probs[probs!=probs] = 0.5 plt.hist(probs, 40) filenames = [os.path.basename(f)for f in filenames] submission = pd.DataFrame({'filename': filenames, 'label': probs}) submission.to_csv('submission.csv', index=False) submission
Deepfake Detection Challenge
8,474,447
<SOS> metric: LogLoss Kaggle data source: deepfake-detection-challenge<data_type_conversions>
from fastai.vision import *
Deepfake Detection Challenge
8,474,447
for col in basement_num_cols: train[col] = train[col].fillna(0) test[col] = test[col].fillna(0 )<define_variables>
train_sample_metadata = pd.read_json('.. /input/deepfake-detection-challenge/train_sample_videos/metadata.json' ).T.reset_index() train_sample_metadata.columns = ['fname','label','split','original'] train_sample_metadata.head()
Deepfake Detection Challenge
8,474,447
masvnr_colums = [] regexp = re.compile(r"([-a-zA-Z]+)?"+r"MasVnr"+r"([-a-zA-Z]+)?") for col in train.columns: if regexp.search(col): masvnr_colums.append(col) masvnr_colums = np.array(masvnr_colums) masvnr_colums<filter>
fake_sample_df = train_sample_metadata[train_sample_metadata.label == 'FAKE'] real_sample_df = train_sample_metadata[train_sample_metadata.label == 'REAL']
Deepfake Detection Challenge
8,474,447
all(train.loc[train['MasVnrArea'].isnull() ].index == train.loc[train['MasVnrType'].isnull() ].index )<data_type_conversions>
train_dir = Path('/kaggle/input/deepfake-detection-challenge/train_sample_videos/') test_dir = Path('/kaggle/input/deepfake-detection-challenge/test_videos/') train_video_files = get_files(train_dir, extensions=['.mp4']) test_video_files = get_files(test_dir, extensions=['.mp4'] )
Deepfake Detection Challenge
8,474,447
train['MasVnrArea'] = train['MasVnrArea'].fillna(0) test['MasVnrArea'] = test['MasVnrArea'].fillna(0 )<data_type_conversions>
dummy_video_file = train_video_files[0]
Deepfake Detection Challenge
8,474,447
train['MasVnrType'] = train['MasVnrType'].fillna('None') test['MasVnrType'] = test['MasVnrType'].fillna('None' )<define_variables>
sys.path.insert(0,'/kaggle/working/reader/python') set_bridge('torch') device = torch.device("cuda" )
Deepfake Detection Challenge
8,474,447
pool_colums = [] regexp = re.compile(r"([-a-zA-Z]+)?"+r"Pool"+r"([-a-zA-Z]+)?") for col in train.columns: if regexp.search(col): pool_colums.append(col) pool_colums = np.array(pool_colums) pool_colums<data_type_conversions>
retinaface_stats = tensor([123,117,104] ).to(device) def decord_cpu_video_reader(path, freq=None): video = VideoReader(str(path), ctx=cpu()) len_video = len(video) if freq: t = video.get_batch(range(0, len(video), freq)).permute(0,3,1,2) else: t = video.get_batch(range(len_video)) return t, len_video def get_decord_video_batch_cpu(path, freq=10, sz=640, stats:Tensor=None, device=defaults.device): "get resized and mean substracted batch tensor of a sampled video(scale of 255)" t_raw, len_video = decord_cpu_video_reader(path, freq) H,W = t_raw.size(2), t_raw.size(3) t = F.interpolate(t_raw.to(device ).to(torch.float32),(sz,sz)) if stats is not None: t -= stats[...,None,None] return t, t_raw,(H, W )
Deepfake Detection Challenge
8,474,447
train['PoolQC'] = train['PoolQC'].fillna('NA') test['PoolQC'] = test['PoolQC'].fillna('NA' )<data_type_conversions>
sys.path.insert(0,"/kaggle/input/retina-face-2/Pytorch_Retinaface_2/" )
Deepfake Detection Challenge
8,474,447
train['Alley'] = train['Alley'].fillna('NA') test['Alley'] = test['Alley'].fillna('NA' )<define_variables>
import os import torch import torch.backends.cudnn as cudnn import numpy as np from data import cfg_mnet, cfg_re50 from layers.functions.prior_box import PriorBox from utils.nms.py_cpu_nms import py_cpu_nms import cv2 from models.retinaface import RetinaFace from utils.box_utils import decode, decode_landm import time
Deepfake Detection Challenge
8,474,447
fireplace_colums = [] regexp = re.compile(r"([-a-zA-Z]+)?"+r"Fireplace"+r"([-a-zA-Z]+)?") for col in train.columns: if regexp.search(col): fireplace_colums.append(col) fireplace_colums = np.array(fireplace_colums) fireplace_colums<data_type_conversions>
def check_keys(model, pretrained_state_dict): ckpt_keys = set(pretrained_state_dict.keys()) model_keys = set(model.state_dict().keys()) used_pretrained_keys = model_keys & ckpt_keys unused_pretrained_keys = ckpt_keys - model_keys missing_keys = model_keys - ckpt_keys print('Missing keys:{}'.format(len(missing_keys))) print('Unused checkpoint keys:{}'.format(len(unused_pretrained_keys))) print('Used keys:{}'.format(len(used_pretrained_keys))) assert len(used_pretrained_keys)> 0, 'load NONE from pretrained checkpoint' return True def remove_prefix(state_dict, prefix): print('remove prefix '{}''.format(prefix)) f = lambda x: x.split(prefix, 1)[-1] if x.startswith(prefix)else x return {f(key): value for key, value in state_dict.items() } def load_model(model, pretrained_path, load_to_cpu): print('Loading pretrained model from {}'.format(pretrained_path)) if load_to_cpu: pretrained_dict = torch.load(pretrained_path, map_location=lambda storage, loc: storage) else: device = torch.cuda.current_device() pretrained_dict = torch.load(pretrained_path, map_location=lambda storage, loc: storage.cuda(device)) if "state_dict" in pretrained_dict.keys() : pretrained_dict = remove_prefix(pretrained_dict['state_dict'], 'module.') else: pretrained_dict = remove_prefix(pretrained_dict, 'module.') check_keys(model, pretrained_dict) model.load_state_dict(pretrained_dict, strict=False) return model
Deepfake Detection Challenge
8,474,447
train['FireplaceQu'] = train['FireplaceQu'].fillna('NA') test['FireplaceQu'] = test['FireplaceQu'].fillna('NA' )<data_type_conversions>
cudnn.benchmark = True
Deepfake Detection Challenge
8,474,447
train['Fence'] = train['Fence'].fillna('NA') test['Fence'] = test['Fence'].fillna('NA' )<define_variables>
def get_model(modelname="mobilenet"): torch.set_grad_enabled(False) cfg = None cfg_mnet['pretrain'] = False cfg_re50['pretrain'] = False if modelname == "mobilenet": pretrained_path = ".. /input/retina-face-2/Pytorch_Retinaface_2/weights/mobilenet0.25_Final.pth" cfg = cfg_mnet if modelname == "resnet50": pretrained_path = ".. /input/retina-face/Pytorch_Retinaface/weights/Resnet50_Final.pth" cfg = cfg_re50 net = RetinaFace(cfg=cfg, phase='test') net = load_model(net, pretrained_path, False) net.eval().to(device) return net, cfg
Deepfake Detection Challenge
8,474,447
misc_colums = [] regexp = re.compile(r"([-a-zA-Z]+)?"+r"Misc"+r"([-a-zA-Z]+)?") for col in train.columns: if regexp.search(col): misc_colums.append(col) misc_colums = np.array(misc_colums) misc_colums<data_type_conversions>
def predict(model, t, sz, cfg, confidence_threshold = 0.5, top_k = 5, nms_threshold = 0.5, keep_top_k = 5): "get prediction for a batch t by model with image sz" resize = 1 scale_rate = 1 im_height, im_width = sz, sz scale = torch.Tensor([sz, sz, sz, sz]) scale = scale.to(device) locs, confs, landmss = torch.Tensor([]), torch.Tensor([]), torch.Tensor([]) locs = locs.to(device) confs = confs.to(device) landmss = landmss.to(device) locs_, confs_, landmss_ = model(t) locs = torch.cat(( locs, locs_), 0) confs = torch.cat(( confs, confs_), 0) landmss = torch.cat(( landmss, landmss_), 0) bbox_result, landms_result = [], [] priorbox = PriorBox(cfg, image_size=(im_height, im_width)) priors = priorbox.forward() priors = priors.to(device) prior_data = priors.data for idx in range(t.size(0)) : loc = locs[idx] conf = confs[idx] landms = landmss[idx] boxes = decode(loc.data.squeeze(0), prior_data, cfg['variance']) boxes = boxes * scale / resize boxes = boxes.cpu().numpy() scores = conf.squeeze(0 ).data.cpu().numpy() [:, 1] landms = decode_landm(landms.data.squeeze(0), prior_data, cfg['variance']) scale1 = torch.Tensor([t.shape[3], t.shape[2], t.shape[3], t.shape[2], t.shape[3], t.shape[2], t.shape[3], t.shape[2], t.shape[3], t.shape[2]]) scale1 = scale1.to(device) landms = landms * scale1 / resize landms = landms.cpu().numpy() inds = np.where(scores > confidence_threshold)[0] boxes = boxes[inds] landms = landms[inds] scores = scores[inds] order = scores.argsort() [::-1][:top_k] boxes = boxes[order] landms = landms[order] scores = scores[order] dets = np.hstack(( boxes, scores[:, np.newaxis])).astype(np.float32, copy=False) keep = py_cpu_nms(dets, nms_threshold) dets = dets[keep, :] landms = landms[keep] dets = dets[:keep_top_k, :] landms = landms[:keep_top_k, :] bbox_result.append(dets[:,:-1]) landms_result.append(landms) return bbox_result, landms_result
Deepfake Detection Challenge
8,474,447
train['MiscFeature'] = train['MiscFeature'].fillna('NA') test['MiscFeature'] = test['MiscFeature'].fillna('NA' )<drop_column>
%%time model, cfg = get_model("mobilenet" )
Deepfake Detection Challenge
8,474,447
train = train.drop(train.loc[train['Electrical'].isnull() ].index )<feature_engineering>
def bboxes_to_original_scale(bboxes, H, W, sz): res = [] for bb in bboxes: h_scale, w_scale = H/sz, W/sz orig_bboxes =(bb*array([w_scale, h_scale, w_scale, h_scale])[None,...] ).astype(int) res.append(orig_bboxes) return res
Deepfake Detection Challenge
8,474,447
for i in test['Neighborhood'].unique() : if test.MSZoning[test['Neighborhood'] == i].isnull().sum() > 0: test.loc[test['Neighborhood'] == i,'MSZoning'] = \ test.loc[test['Neighborhood'] == i,'MSZoning'].fillna(test.loc[test['Neighborhood'] == i,'MSZoning'].mode() [0] )<data_type_conversions>
def landmarks_to_original_scale(landmarks, H, W, sz): res = [] for landms in landmarks: h_scale, w_scale = H/sz, W/sz orig_landms =(landms*array([w_scale, h_scale]*5)[None,...] ).astype(int) res.append(orig_landms) return res
Deepfake Detection Challenge
8,474,447
train['LotFrontage'].fillna(train['LotFrontage'].median() , inplace=True) test['LotFrontage'].fillna(test['LotFrontage'].median() , inplace=True )<prepare_x_and_y>
from tqdm import tqdm
Deepfake Detection Challenge
8,474,447
y_train = train['SalePrice'] train = train.drop('SalePrice',axis=1) train = train.drop('Id',axis=1 )<drop_column>
freq = 5 model_args = dict(confidence_threshold = 0.5, top_k = 5, nms_threshold = 0.5, keep_top_k = 5) sz = cfg['image_size'] imgnet_stats = [tensor(o)for o in imagenet_stats] rescale_param = 1.3
Deepfake Detection Challenge
8,474,447
test_id = test['Id'] test = test.drop('Id', axis=1 )<concatenate>
!pip install -q.. /input/efficientnetpytorchpip/efficientnet_pytorch-0.6.3/
Deepfake Detection Challenge
8,474,447
df = train.append(test )<categorify>
from fastai.vision.models.efficientnet import *
Deepfake Detection Challenge
8,474,447
df['GarageCond'] = df['GarageCond'].map({'NA':0, 'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['GarageQual'] = df['GarageQual'].map({'NA':0, 'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['BsmtCond'] = df['BsmtCond'].map({'NA':0, 'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['BsmtExposure'] = df['BsmtExposure'].map({'NA':0, 'No':1, 'Mn':2, 'Av':3, 'Gd':4}) df['BsmtFinType1'] = df['BsmtFinType1'].map({'NA':0, 'Unf':1, 'LwQ':2, 'Rec':3, 'BLQ':4, 'ALQ':5, 'GLQ':6}) df['BsmtFinType2'] = df['BsmtFinType2'].map({'NA':0, 'Unf':1, 'LwQ':2, 'Rec':3, 'BLQ':4, 'ALQ':5, 'GLQ':6}) df['BsmtQual'] = df['BsmtQual'].map({'NA':0, 'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['PoolQC'] = df['PoolQC'].map({'NA':0, 'Fa':1, 'TA':2, 'Gd':3, 'Ex':4}) df['Alley'] = df['Alley'].map({'NA':0, 'Grvl':1, 'Pave':2}) df['FireplaceQu'] = df['FireplaceQu'].map({'NA':0, 'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['ExterCond'] = df['ExterCond'].map({'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['ExterQual'] = df['ExterQual'].map({'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['KitchenQual'] = df['KitchenQual'].map({'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['LandSlope'] = df['LandSlope'].map({'Sev':1, 'Mod':2, 'Gtl':3}) df['PavedDrive'] = df['PavedDrive'].map({'N':1, 'P':2, 'Y':3}) df['Functional'] = df['Functional'].map({'Sal':1, 'Sev':2, 'Maj2':3, 'Maj1':4, 'Mod':5, 'Min2':6, 'Min1':7, 'Typ':8}) df['HeatingQC'] = df['HeatingQC'].map({'Po':1, 'Fa':2, 'TA':3, 'Gd':4, 'Ex':5}) df['Street'] = df['Street'].map({'Grvl':1, 'Pave':2}) df['Utilities'] = df['Utilities'].map({'ELO':1, 'NoSeWa':2, 'NoSewr':3, 'AllPub':4}) df=df.drop('MoSold',axis=1) df['MSSubClass'] = df['MSSubClass'].map({20:'class1', 30:'class2', 40:'class3', 45:'class4', 50:'class5', 60:'class6', 70:'class7', 75:'class8', 80:'class9', 85:'class10', 90:'class11', 120:'class12', 150:'class13', 160:'class14', 180:'class15', 190:'class16'} )<drop_column>
class DummyDatabunch: c = 2 path = '.' device = defaults.device loss_func = None data = DummyDatabunch()
Deepfake Detection Challenge
8,474,447
df = df.reset_index() df = df.drop('index',axis = 1 )<filter>
effnet_model = EfficientNet.from_name("efficientnet-b5", override_params={'num_classes': 2}) learner = Learner(data, effnet_model); learner.model_dir = '.' learner.load('.. /input/deepfakerandmergeaugmodels/single_frame_effnetb5_randmerge') effnetb5_inference_model = learner.model.eval()
Deepfake Detection Challenge
8,474,447
drop_id = df[df['LotArea'] > 100000].index<drop_column>
effnet_model = EfficientNet.from_name("efficientnet-b7", override_params={'num_classes': 2}) learner = Learner(data, effnet_model); learner.model_dir = '.' learner.load('.. /input/deepfakerandmergeaugmodels/single_frame_effnetb7_randmerge_fp16') effnetb7_inference_model = learner.model.float().eval()
Deepfake Detection Challenge
8,474,447
drop_id = drop_id[drop_id < 1459]<feature_engineering>
learner = cnn_learner(data, models.resnet34, pretrained=False); learner.model_dir = '.' learner.load('.. /input/deepfakerandmergeaugmodels/single_frame_resnet34_randmerge') resnet_inference_model = learner.model.eval()
Deepfake Detection Challenge
8,474,447
df['MasVnrArea'][df[df['MasVnrArea'] > 1500].index] = df['MasVnrArea'].mean() df['Utilities'][df[df['Utilities']==2].index] = df['Utilities'].mean()<drop_column>
predictions = [] video_fnames = []
Deepfake Detection Challenge
8,474,447
df = df.drop(drop_id )<drop_column>
fname2pred = dict(zip(video_fnames, predictions))
Deepfake Detection Challenge
8,474,447
y_train = y_train.drop(drop_id )<categorify>
submission_df = pd.read_csv("/kaggle/input/deepfake-detection-challenge/sample_submission.csv" )
Deepfake Detection Challenge
8,474,447
dummy_drop = [] for i in cat_columns: dummy_drop += [ i+'_'+str(df[i].unique() [-1])] df = pd.get_dummies(df,columns=cat_columns) df = df.drop(dummy_drop,axis=1 )<normalization>
submission_df.label = submission_df.filename.map(fname2pred )
Deepfake Detection Challenge
8,474,447
X_train = df[:-1459].drop(['index'], axis=1) X_test = df[-1459:].drop(['index'], axis=1) scaler = StandardScaler() X_train[num_columns]= scaler.fit_transform(X_train[num_columns]) X_test[num_columns]= scaler.transform(X_test[num_columns]) X_train.shape, X_test.shape<feature_engineering>
submission_df['label'] = np.clip(submission_df['label'], 0.01, 0.99 )
Deepfake Detection Challenge
8,474,447
<train_model><EOS>
submission_df.to_csv("submission.csv",index=False )
Deepfake Detection Challenge
8,212,093
<SOS> metric: LogLoss Kaggle data source: deepfake-detection-challenge<train_on_grid>
%matplotlib inline
Deepfake Detection Challenge
8,212,093
ans = {} for i in range(1, 222): imp_col = imp_feature.iloc[:i].index ridge = KernelRidge(alpha = 0.5263157894736842, coef0 = 3.5, degree = 2, kernel ='polynomial') ridge = ridge.fit(X_train[imp_col], y_train_log) ans[i] = np.sqrt(mean_squared_error(y_train_log,ridge.predict(X_train[imp_col])) )<define_variables>
test_dir = "/kaggle/input/deepfake-detection-challenge/test_videos/" test_videos = sorted([x for x in os.listdir(test_dir)if x[-4:] == ".mp4"]) len(test_videos )
Deepfake Detection Challenge
8,212,093
minimum = ans[1] ind_min = 1 for ind in range(1,len(ans.values())) : if ans[ind] < minimum: minimum = ans[ind] ind_min = ind<features_selection>
print("PyTorch version:", torch.__version__) print("CUDA version:", torch.version.cuda) print("cuDNN version:", torch.backends.cudnn.version() )
Deepfake Detection Challenge
8,212,093
imp_col = imp_feature.iloc[:ind_min+1].index<compute_test_metric>
gpu = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") gpu
Deepfake Detection Challenge
8,212,093
def neg_rmse(y_true, y_pred): return -1.0*np.sqrt(mean_squared_error(y_true,y_pred)) neg_rmse = make_scorer(neg_rmse )<train_model>
facedet = BlazeFace().to(gpu) facedet.load_weights("/kaggle/input/blazeface-pytorch/blazeface.pth") facedet.load_anchors("/kaggle/input/blazeface-pytorch/anchors.npy") _ = facedet.train(False )
Deepfake Detection Challenge
8,212,093
model = KernelRidge(alpha = 0.6842105263157894, coef0 = 3.5, degree = 2, kernel = 'polynomial') model.fit(X_train[imp_col], y_train_log) print("RMSE of the whole training set: {}".format(np.sqrt(mean_squared_error(y_train_log,model.predict(X_train[imp_col])))) )<predict_on_test>
frames_per_video = 17 video_reader = VideoReader() video_read_fn = lambda x: video_reader.read_frames(x, num_frames=frames_per_video) face_extractor = FaceExtractor(video_read_fn, facedet )
Deepfake Detection Challenge
8,212,093
y_pred = np.exp(model.predict(X_test[imp_col]))<save_to_csv>
input_size = 224
Deepfake Detection Challenge
8,212,093
def save_ans(ans, pasanger_id, name_alg): submission = pd.DataFrame({'Id':pasanger_id,'SalePrice':ans}) print(submission.shape) filename = r'./{}.csv'.format(name_alg) submission.to_csv(filename,index=False) print('Saved file: ' + filename )<compute_test_metric>
mean = [0.485, 0.456, 0.406] std = [0.229, 0.224, 0.225] normalize_transform = Normalize(mean, std )
Deepfake Detection Challenge
8,212,093
save_ans(y_pred, test_id,'submission' )<load_from_csv>
class MyResNeXt(models.resnet.ResNet): def __init__(self, training=True): super(MyResNeXt, self ).__init__(block=models.resnet.Bottleneck, layers=[3, 4, 6, 3], groups=32, width_per_group=4) self.fc = nn.Linear(2048, 1 )
Deepfake Detection Challenge
8,212,093
train_df = pd.read_csv('.. /input/house-prices-advanced-regression-techniques/train.csv') test_df = pd.read_csv('.. /input/house-prices-advanced-regression-techniques/test.csv' )<drop_column>
checkpoint = torch.load("/kaggle/input/deepfakes-inference-demo/resnext.pth", map_location=gpu) model = MyResNeXt().to(gpu) model.load_state_dict(checkpoint) _ = model.eval() del checkpoint
Deepfake Detection Challenge
8,212,093
def preprocessing_null(data_df): data_df.drop(['Alley', 'PoolQC', 'Fence', 'MiscFeature', 'Id'], axis=1, inplace=True) Bsmtlist = ['BsmtQual', 'BsmtCond', 'BsmtExposure', 'BsmtFinType1', 'BsmtFinType2'] Bsmtlist2=['BsmtFinSF1', 'BsmtFinSF2', 'BsmtUnfSF'] Garagelist = ['GarageType', 'GarageFinish', 'GarageQual', 'GarageCond'] Bathlist = ['BsmtFullBath', 'BsmtHalfBath'] Extlist = ['Exterior1st', 'Exterior2nd'] data_df.loc[:, Bsmtlist]=data_df.loc[:, Bsmtlist].fillna('TA') data_df['Electrical']=data_df['Electrical'].fillna('SBrkr') data_df['LotFrontage']=data_df['LotFrontage'].fillna(data_df['LotFrontage'].mean()) data_df['FireplaceQu'] = data_df['FireplaceQu'].fillna('NA') data_df.loc[:, Garagelist] = data_df.loc[:, Garagelist].fillna('NA') data_df['GarageYrBlt']=data_df['GarageYrBlt'].fillna(2005) data_df.loc[:, 'MasVnrType'] = data_df.loc[:, 'MasVnrType'].fillna('None') data_df['MasVnrArea']=data_df['MasVnrArea'].fillna(0) data_df.loc[:, Bsmtlist2]=data_df[Bsmtlist2].fillna(0) data_df['TotalBsmtSF']=data_df['TotalBsmtSF'].fillna(0) data_df['GarageArea']=data_df['GarageArea'].fillna(data_df['GarageArea'].median()) data_df['GarageCars']=data_df['GarageCars'].fillna(data_df['GarageCars'].median()) data_df[Bathlist]=data_df[Bathlist].fillna(0) data_df[Extlist]=data_df[Extlist].fillna('VinylSd') data_df['MSZoning']=data_df['MSZoning'].fillna('TA') data_df['Utilities']=data_df['Utilities'].fillna('AllPub') data_df['KitchenQual']=data_df['KitchenQual'].fillna('TA') data_df['Functional']=data_df['Functional'].fillna('Typ') data_df['SaleType']=data_df['SaleType'].fillna('WD') return data_df<prepare_x_and_y>
def predict_on_video(video_path, batch_size): try: faces = face_extractor.process_video(video_path) face_extractor.keep_only_best_face(faces) if len(faces)> 0: x = np.zeros(( batch_size, input_size, input_size, 3), dtype=np.uint8) n = 0 for frame_data in faces: for face in frame_data["faces"]: resized_face = isotropically_resize_image(face, input_size) resized_face = make_square_image(resized_face) if n < batch_size: x[n] = resized_face n += 1 else: print("WARNING: have %d faces but batch size is %d" %(n, batch_size)) if n > 0: x = torch.tensor(x, device=gpu ).float() x = x.permute(( 0, 3, 1, 2)) for i in range(len(x)) : x[i] = normalize_transform(x[i] / 255.) with torch.no_grad() : y_pred = model(x) y_pred = torch.sigmoid(y_pred.squeeze()) return y_pred[:n].mean().item() except Exception as e: print("Prediction error on video %s: %s" %(video_path, str(e))) return 0.5
Deepfake Detection Challenge
8,212,093
train_target = train_df['SalePrice'] train_feature = train_df.drop('SalePrice', axis=1 )<categorify>
def predict_on_video_set(videos, num_workers): def process_file(i): filename = videos[i] y_pred = predict_on_video(os.path.join(test_dir, filename), batch_size=frames_per_video) return y_pred with ThreadPoolExecutor(max_workers=num_workers)as ex: predictions = ex.map(process_file, range(len(videos))) return list(predictions )
Deepfake Detection Challenge
8,212,093
train_feature = preprocessing_null(train_feature) test_feature = preprocessing_null(test_df )<set_options>
speed_test = False
Deepfake Detection Challenge
8,212,093
%matplotlib inline<set_options>
if speed_test: start_time = time.time() speedtest_videos = test_videos[:5] predictions = predict_on_video_set(speedtest_videos, num_workers=4) elapsed = time.time() - start_time print("Elapsed %f sec.Average per video: %f sec." %(elapsed, elapsed / len(speedtest_videos)) )
Deepfake Detection Challenge
8,212,093
pd.set_option('display.max_columns', 500) corr[corr>0.7]<drop_column>
predictions = predict_on_video_set(test_videos, num_workers=4 )
Deepfake Detection Challenge
8,212,093
<split><EOS>
submission_df = pd.DataFrame({"filename": test_videos, "label": predictions}) submission_df.to_csv("submission.csv", index=False )
Deepfake Detection Challenge
8,069,381
<SOS> metric: LogLoss Kaggle data source: deepfake-detection-challenge<split>
%matplotlib inline
Deepfake Detection Challenge
8,069,381
train_feature_num, train_feature_obj = split_num_obj(train_feature) test_feature_num, test_feature_obj = split_num_obj(test_feature )<categorify>
test_dir = "/kaggle/input/deepfake-detection-challenge/test_videos/" test_videos = sorted([x for x in os.listdir(test_dir)if x[-4:] == ".mp4"]) frame_h = 5 frame_l = 5 len(test_videos )
Deepfake Detection Challenge
8,069,381
train_dummies = pd.get_dummies(train_feature_obj) test_dummies = pd.get_dummies(test_feature_obj) not_in_train = [column for column in train_dummies.columns if column not in test_dummies.columns] not_in_test = [column for column in test_dummies.columns if column not in train_dummies.columns] print(' ',train_dummies.shape, test_dummies.shape, ' ') print(' ', not_in_train, ' ') print(' ', not_in_test )<define_search_space>
print("PyTorch version:", torch.__version__) print("CUDA version:", torch.version.cuda) print("cuDNN version:", torch.backends.cudnn.version() )
Deepfake Detection Challenge
8,069,381
df_num_cat_col = df_num_col[[0, 3, 4, 15, 16, 17, 18, 19, 20, 21, 22, 28, 29, 30, 31]]<split>
gpu = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") gpu
Deepfake Detection Challenge
8,069,381
def split_num_cat(data_df_num): data_df_num_cat = data_df_num[df_num_cat_col] data_df_num_non_cat = data_df_num.drop(df_num_cat_col, axis=1) return data_df_num_cat, data_df_num_non_cat train_feature_num_cat, train_feature_num_non_cat = split_num_cat(train_feature_num) test_feature_num_cat, test_feature_num_non_cat = split_num_cat(test_feature_num) print(train_feature_num_cat.shape,test_feature_num_cat.shape) print(train_feature_num_non_cat.shape, test_feature_num_non_cat.shape )<split>
facedet = BlazeFace().to(gpu) facedet.load_weights("/kaggle/input/blazeface-pytorch/blazeface.pth") facedet.load_anchors("/kaggle/input/blazeface-pytorch/anchors.npy") _ = facedet.train(False )
Deepfake Detection Challenge
8,069,381
all_data = pd.concat(( train_df, test_df)).reset_index(drop=True) all_data.drop('SalePrice', axis=1, inplace=True) print(all_data.shape) all_data = preprocessing_null(all_data) all_data = drop_corr_ftr(all_data) print(all_data.shape) all_data_num, all_data_obj = split_num_obj(all_data) print(all_data_obj.shape )<categorify>
frames_per_video = 64 video_reader = VideoReader() video_read_fn = lambda x: video_reader.read_frames(x, num_frames=frames_per_video) face_extractor = FaceExtractor(video_read_fn, facedet )
Deepfake Detection Challenge
8,069,381
warnings.filterwarnings(action='ignore') label = LabelEncoder() for col in all_data_obj.columns: label.fit(all_data_obj.loc[:, col]) train_feature_obj.loc[:, col] = label.transform(train_feature_obj.loc[:, col]) test_feature_obj.loc[:, col] = label.transform(test_feature_obj.loc[:, col] )<feature_engineering>
input_size = 224
Deepfake Detection Challenge
8,069,381
train_target = np.log1p(train_target) train_feature_num_non_cat = np.log1p(train_feature_num_non_cat) test_feature_num_non_cat = np.log1p(test_feature_num_non_cat )<concatenate>
mean = [0.485, 0.456, 0.406] std = [0.229, 0.224, 0.225] normalize_transform = Normalize(mean, std )
Deepfake Detection Challenge
8,069,381
train_feature_fin = pd.concat([train_feature_num_cat, train_feature_num_non_cat, train_feature_obj], axis=1) test_feature_fin = pd.concat([test_feature_num_cat, test_feature_num_non_cat, test_feature_obj], axis=1 )<define_variables>
class MyResNeXt(models.resnet.ResNet): def __init__(self, training=True): super(MyResNeXt, self ).__init__(block=models.resnet.Bottleneck, layers=[3, 4, 6, 3], groups=32, width_per_group=4) self.fc = nn.Linear(2048, 1 )
Deepfake Detection Challenge
8,069,381
cond1 = train_target>500000 cond2 = train_feature_num_non_cat['GrLivArea']>4000<drop_column>
checkpoint = torch.load("/kaggle/input/deepfakes-inference-demo/resnext.pth", map_location=gpu) model = MyResNeXt().to(gpu) model.load_state_dict(checkpoint) _ = model.eval() del checkpoint
Deepfake Detection Challenge
8,069,381
train_feature_fin = train_feature_fin.drop(train_feature_num_non_cat[cond1|cond2].index, axis=0 )<train_on_grid>
def predict_on_video(video_path, batch_size): try: faces = face_extractor.process_video(video_path) face_extractor.keep_only_best_face(faces) if len(faces)> 0: x = np.zeros(( batch_size, input_size, input_size, 3), dtype=np.uint8) n = 0 for frame_data in faces: for face in frame_data["faces"]: resized_face = isotropically_resize_image(face, input_size) resized_face = make_square_image(resized_face) if n < batch_size: x[n] = resized_face n += 1 else: print("WARNING: have %d faces but batch size is %d" %(n, batch_size)) if n > 0: x = torch.tensor(x, device=gpu ).float() x = x.permute(( 0, 3, 1, 2)) for i in range(len(x)) : x[i] = normalize_transform(x[i] / 255.) with torch.no_grad() : y_pred = model(x) y_pred = torch.sigmoid(y_pred.squeeze()) return y_pred[:n].mean().item() except Exception as e: print("Prediction error on video %s: %s" %(video_path, str(e))) return 0.481
Deepfake Detection Challenge
8,069,381
def get_best_estimator(model, params): grid_model = GridSearchCV(model, param_grid=params, scoring="neg_mean_squared_error", cv=5) grid_model.fit(train_feature_fin, train_target) rmse = np.sqrt(-1*grid_model.best_score_) print('{0}, param:{1}, rmse:{2}'.format(model.__class__.__name__, grid_model.best_params_,\ np.round(rmse, 4))) return grid_model.best_estimator_ ridge_params = {'alpha':[0.05, 0.1, 1, 5, 8, 10, 15]} lasso_params = {'alpha':[0.001, 0.005, 0.008, 0.05, 0.1, 0.3, 0.5, 1, 5, 10]} elastic_params = {'alpha':[0.05, 0.1, 0.5, 1, 3, 5, 8]} ridge_reg = Ridge() lasso_reg = Lasso() elastic_reg = ElasticNet(l1_ratio=0.7) lasso_be = get_best_estimator(lasso_reg, lasso_params) ridge_be = get_best_estimator(ridge_reg, ridge_params) elastic_be = get_best_estimator(elastic_reg, elastic_params )<choose_model_class>
def predict_on_video_set(videos, num_workers): def process_file(i): filename = videos[i] y_pred = predict_on_video(os.path.join(test_dir, filename), batch_size=frames_per_video) return y_pred with ThreadPoolExecutor(max_workers=num_workers)as ex: predictions = ex.map(process_file, range(len(videos))) return list(predictions )
Deepfake Detection Challenge
8,069,381
lgbm_params = { 'max_depth':[5, 10, 15, 20, 25, 30], 'learning_rate':[0.01, 0.05, 0.1, 0.5, 1], } lgbm_reg = LGBMRegressor(n_estimators=1000) lgbm_be = get_best_estimator(lgbm_reg, lgbm_params )<predict_on_test>
predictions = predict_on_video_set(test_videos, num_workers=4 )
Deepfake Detection Challenge
8,069,381
preds = np.expm1(lgbm_be.predict(test_feature_fin))<save_to_csv>
submission_df_resnext = pd.DataFrame({"filename": test_videos, "label": predictions}) submission_df_resnext.to_csv("submission_resnext.csv", index=False )
Deepfake Detection Challenge
8,069,381
test=pd.read_csv('.. /input/house-prices-advanced-regression-techniques/test.csv') my_submission = pd.DataFrame({'Id': test.Id, 'SalePrice': preds}) my_submission.to_csv('submission.csv', index=False )<load_from_csv>
!pip install.. /input/deepfake-xception-trained-model/pytorchcv-0.0.55-py2.py3-none-any.whl --quiet
Deepfake Detection Challenge
8,069,381
train=pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/train.csv',index_col=0) test=pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/test.csv',index_col=0) submission=pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/sample_submission.csv') Id_test = pd.read_csv('/kaggle/input/house-prices-advanced-regression-techniques/test.csv' ).values[:,0] data=train print(train.shape, test.shape, submission.shape, Id_test.shape )<set_options>
%matplotlib inline warnings.filterwarnings("ignore" )
Deepfake Detection Challenge
8,069,381
%matplotlib inline <drop_column>
test_dir = "/kaggle/input/deepfake-detection-challenge/test_videos/" test_videos = sorted([x for x in os.listdir(test_dir)if x[-4:] == ".mp4"]) len(test_videos )
Deepfake Detection Challenge