def __init__(self, model_name, num_classes=5, pretrained=False,
aux_logits=False, freeze_features=False):
super(DR_Classifier, self).__init__()
self.model_name = model_name
self.num_classes = num_classes
self.aux_logits = aux_logits
if self.model_name == 'inception-v3':
self.model = models.inception_v3(pretrained=pretrained,
aux_logits=aux_logits)
num_ftrs = self.model.fc.in_features
self.model.fc = self.createLayer(num_ftrs)
if self.aux_logits:
num_ftrs = self.model.AuxLogits.fc.in_features
self.model.AuxLogits.fc = self.createLayer(num_ftrs)
elif self.model_name == "densenet-161":
self.model = models.densenet161(pretrained=pretrained)
num_ftrs = self.model.classifier.in_features
self.model.classifier = self.createLayer(num_ftrs)
elif self.model_name == "efficientnet-b7":
self.model = EfficientNet.from_pretrained(model_name)
num_ftrs = self.model._fc.in_features
self.model._fc = self.createLayer(num_ftrs)
elif self.model_name == "efficientnet-b5":
self.model = EfficientNet.from_pretrained(model_name)
num_ftrs = self.model._fc.in_features
self.model._fc = self.createLayer(num_ftrs)
elif self.model_name == "wideresnet":
self.model = models.wide_resnet101_2(pretrained = True)
num_ftrs = self.model.fc.in_features
self.model.fc = self.createLayer(num_ftrs)
else:
print("Invalid model name, exiting...")
return;
def feature_wide_resnet101(pretrained=True, **kwargs):
kwargs['width_per_group'] = 64 * 2
model = FeatureResNet(Bottleneck, [3, 4, 23, 3], **kwargs)
if pretrained:
model.load_state_dict(
models.wide_resnet101_2(pretrained=True).state_dict())
return model
def wide_resnet101_2(pretrained: bool,
progress: bool = True,
requires_grad: bool = True):
model = models.wide_resnet101_2(pretrained=pretrained, progress=progress)
for params in model.parameters():
params.requires_grad = requires_grad
return model
def __init__(self, args: Namespace) -> None:
# Save parameters
self.data_dir = args.data_dir
self.tsvname = args.tsvname
self.embed_dir = args.embed_dir
self.timeout = args.timeout
self.log_every = args.log_every
# Turn PIL warnings into exceptions to filter out bad images
simplefilter('error', Image.DecompressionBombWarning)
simplefilter('error', UserWarning)
# Automatically use GPU if available
if not cuda.is_available():
raise RuntimeError(
'Must have CUDA installed in order to run this program.')
self.device = device('cuda')
self.model = wide_resnet101_2(pretrained=True, progress=True)
# Move model to device
self.model.to(self.device)
self.model.eval() # Don't forget to put model in evaluation mode!
# Use recommended sequence of transforms for ImageNet pretrained models
self.transforms = Compose([Resize(256, interpolation=Image.BICUBIC), # Default is bilinear
CenterCrop(224),
ToTensor(),
Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])])
self.embeddings = Queue(self.log_every) # Thread-safe
self.model.avgpool.register_forward_hook(lambda m, m_in, m_out: self.embeddings.put(
m_out.data.detach().cpu().squeeze().numpy()))
def get_model(name, pretrained=True):
if name == 'resnet18':
model = models.resnet18(pretrained=pretrained)
out_channels = [64, 128, 256, 512]
elif name == 'resnet34':
model = models.resnet34(pretrained=pretrained)
out_channels = [64, 128, 256, 512]
elif name == 'resnet50':
model = models.resnet50(pretrained=pretrained)
out_channels = [256, 512, 1024, 2048]
elif name == 'resnet101':
model = models.resnet101(pretrained=pretrained)
out_channels = [256, 512, 1024, 2048]
# ResNeXt
elif name == 'resnext50':
model = models.resnext50_32x4d(pretrained=pretrained)
out_channels = [256, 512, 1024, 2048]
elif name == 'resnext101':
model = models.resnext101_32x8d(pretrained=pretrained)
out_channels = [256, 512, 1024, 2048]
elif name == 'wide_resnet50':
model = models.wide_resnet50_2(pretrained=pretrained)
out_channels = [256, 512, 1024, 2048]
elif name == 'wide_resnet101':
model = models.wide_resnet101_2(pretrained=pretrained)
out_channels = [256, 512, 1024, 2048]
# Error
else:
raise NotImplemented('{} backbone model is not implemented so far.'.format(name))
return model, out_channels
def __init__(self, c_in, model_name):
super(PretrainedNet, self).__init__()
if model_name == 'resnet101':
model_ft = models.resnet101(pretrained=True)
elif model_name == 'densenet121':
model_ft = models.densenet121(pretrained=True)
elif model_name == 'densenet161':
model_ft = models.densenet161(pretrained=True)
elif model_name == 'squeezenet':
model_ft = models.squeezenet1_0(pretrained=True)
elif model_name == 'resnext101_32x8d':
model_ft = models.wide_resnet101_2(pretrained=True)
elif model_name == 'inception_v3':
model_ft = models.inception_v3(pretrained=True)
self._set_parameter_requires_grad(model_ft)
if model_name.startswith('res'):
layers = [i for j, i in enumerate(model_ft.children()) if j < 7]
self.model = nn.Sequential(*layers, nn.MaxPool2d((4, 1)))
elif model_name.startswith('dense'):
layers = model_ft.features[:-3]
self.model = nn.Sequential(*layers, nn.MaxPool2d((4, 1)))
elif model_name.startswith('squeeze'):
layers = model_ft.features[:-2]
self.model = nn.Sequential(*layers, nn.MaxPool2d((4, 1)))
elif model_name.startswith('incep'):
layers = [i for j, i in enumerate(model_ft.children()) if j < 13]
self.model = nn.Sequential(*layers, nn.MaxPool2d((8, 4)))
def __init__(self,
device: Union[str, torch.device] = "cpu",
load_from_path: str = None,
augmentations: nn.Sequential = None) -> None:
"""Initialise model and learner setup."""
self.device = device
self.model = models.wide_resnet101_2(pretrained=False).to(self.device)
self.timestamp = datetime.now().strftime("%m-%d-%Y_%H-%M-%S")
if load_from_path is not None:
print("Loading model...")
state_dict = torch.load(load_from_path)
self.model.load_state_dict(state_dict)
if augmentations is None:
self.learner = BYOL(self.model,
image_size=64,
hidden_layer="avgpool")
if augmentations is not None:
self.learner = BYOL(self.model,
image_size=64,
hidden_layer="avgpool",
augment_fn=augmentations)
self.opt = torch.optim.Adam(self.learner.parameters(),
lr=0.0001,
betas=(0.9, 0.999))
self.loss_history: list[float] = []
def extract_resnet_features():
dataset = MetalAlbumsDataset("D:\\metal_data\\album_id_titles.json",
"D:\\metal_data\\covers_128")
data_loader = DataLoader(dataset,
50,
False,
collate_fn=pack_collate_images)
resnet = torch.nn.Sequential(
*list(models.wide_resnet101_2(
pretrained=True).children())[:-1]).to("cuda")
resnet.eval()
all_features = np.zeros((len(dataset), 2048))
with open('album_resnet.csv', 'w', newline='') as f:
writer = csv.writer(f, delimiter=',', quoting=csv.QUOTE_NONNUMERIC)
index = 0
for i, (images, ids) in enumerate(data_loader):
images = images.to("cuda")
with torch.no_grad():
features = resnet(images).cpu().numpy()
all_features[index:index + len(ids), :] = features.reshape(
len(ids), 2048)
index += len(ids)
for j, album_id in enumerate(ids):
writer.writerow(
[album_id, features[j, :, :].reshape(-1).tolist()])
np.save("album_resnet.npy", all_features)
def __init__(self,num_classes = 60,model_name='resnet50'):
super(VA, self).__init__()
self.num_classes = num_classes
self.conv1 = nn.Conv2d(3, 128, kernel_size=5, stride=2,
bias=False)
self.bn1 = nn.BatchNorm2d(128)
self.relu1 = nn.ReLU(inplace=True)
self.conv2 = nn.Conv2d(128, 128, kernel_size=5, stride=2,
bias=False)
self.bn2 = nn.BatchNorm2d(128)
self.relu2 = nn.ReLU(inplace=True)
self.avepool = nn.MaxPool2d(7)
self.fc = nn.Linear(6272, 6)
if model_name=='resnet50':
self.classifier = models.resnet50(pretrained=True)
elif model_name=='resnext50_32x4d':
self.classifier = models.resnext50_32x4d(pretrained=True)
elif model_name=='resnext101_32x8d':
self.classifier = models.resnext101_32x8d(pretrained=True)
elif model_name=='wide_resnet50_2':
self.classifier = models.wide_resnet50_2(pretrained=True)
elif model_name=='wide_resnet101_2':
self.classifier = models.wide_resnet101_2(pretrained=True)
elif model_name=='xception':
self.classifier = pretrainedmodels.xception(num_classes=1000, pretrained='imagenet')
total_params = 0
for nname, pparameter in self.classifier.named_parameters():
if not pparameter.requires_grad: continue
param = pparameter.numel()
total_params+=param
print(f"Total Trainable Params: {total_params}")
self.init_weight()
def wide_resnet101(pretrained=False, imagenet_weight=False):
"""Constructs a Wide_ResNet-101 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
print('=== Using wide_resnet 101 ===')
model = models.wide_resnet101_2()
if pretrained:
if imagenet_weight:
print('=== use {} as backbone'.format(imagenet_weight))
state_dict = torch.load(imagenet_weight)['state_dict']
state_dict = exchange_weightkey_in_state_dict(state_dict)
model.load_state_dict(state_dict)
else:
print('=== use pytorch default backbone')
model = models.wide_resnet101_2(pretrained=True)
return model
def get_wide_resnet101_2(class_num):
model = models.wide_resnet101_2(pretrained=True)
set_parameter_requires_grad(model)
model.name = 'wide_resnet101_2'
n_inputs = model.fc.in_features
model.fc = nn.Linear(n_inputs, class_num)
return model, 224
def get_model(model_name):
"""
Get specific modified pre-trained models by name
"""
model = None
if model_name == "AlexNet":
model = models.alexnet(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "VGG":
model = models.vgg16(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "ResNet":
model = models.resnet18(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "SqueezNet":
model = models.squeezenet1_1(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "DenseNet":
model = models.densenet121(pretrained=True)
in_features = model.classifier.in_features
model.classifier = nn.Linear(in_features, 26)
elif model_name == "Inception":
model = models.inception_v3(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "GoogleNet":
model = models.googlenet(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "ShuffleNet":
model = models.shufflenet_v2_x1_0(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "MobileNet":
model = models.mobilenet_v2(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "ResNext":
model = models.resnext101_32x8d(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "WResNet":
model = models.wide_resnet101_2(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
elif model_name == "MNASNet":
model = models.mnasnet1_0(pretrained=True)
in_features = model.fc.in_features
model.fc = nn.Linear(in_features, 26)
return model
def __init__(self):
super(feature_extractor, self).__init__()
#resnet152 = models.resnet152(pretrained=True)
wideresnet_101_2 = models.wide_resnet101_2(pretrained = True)
modules = list(wideresnet_101_2.children())[:-1]
#modules = list(resnet152.children())[:-1]
self.net = nn.Sequential(*modules)
self.ft_size = 2048
def create_poseresnet_model(enable_binary=False):
# TODO: Test trained vs. pretrained
# training loop
backbone_layer = models.wide_resnet101_2(pretrained=True)
classification = five_layer_classification(512, 256, 512, 256, 3)
if enable_binary:
classification = five_layer_classification(512, 256, 512, 256, 2)
model = pose_resnet(backbone_layer, classification)
return model
def getValueFromJson(file_json, model_path_load):
global is_cuda, optimizer, train_transform
is_cuda = False
if torch.cuda.is_available():
is_cuda = True
gaugedataset = DigitGaugeDataset(file_json, train_transform)
len_digit = len(gaugedataset)
gauge_data_loader = torch.utils.data.DataLoader(gaugedataset,
batch_size=len_digit,
shuffle=False)
data, target = next(iter(gauge_data_loader))
resnet = models.wide_resnet101_2(pretrained=False)
resnet.fc.out_features = 10
if is_cuda == True:
resnet = resnet.cuda()
if is_cuda == True:
dev = torch.device('cuda')
else:
dev = torch.device('cpu')
checkpoint = torch.load(model_path_load, map_location=dev)
resnet.load_state_dict(checkpoint['model_state_dict'])
# for layer in resnet.classifier.children():
# if (type(layer) == nn.Dropout):
# layer.train(False)
# layer.p = 0
resnet.eval()
# resnet.train()
volatile = True
if is_cuda:
data, target = data.cuda(), target.cuda()
data, target = Variable(data, volatile), Variable(target)
output = resnet(data)
preds = output.data.max(dim=1, keepdim=True)[1]
count_correct = preds.eq(target.data.view_as(preds)).cpu().sum()
list_preds = preds.view(-1).tolist()
print(
f'target:{target.view(-1).tolist()}, preds:{list_preds}, accuracy :{count_correct / len_digit}'
)
npred = int(''.join([str(aa) for aa in list_preds]))
if gaugedataset.dict_json_info['digitFractNo'] > 0:
npred = npred / (10**gaugedataset.dict_json_info['digitFractNo'])
return npred
def get_model(model_name, train_set=None):
if model_name in ["linear", "logistic"]:
batch = train_set[0]
model = LinearRegression(input_dim=batch['images'].shape[0],
output_dim=1)
if model_name == "mlp":
model = Mlp(n_classes=10, dropout=False)
if model_name == "mlp_dropout":
model = Mlp(n_classes=10, dropout=True)
elif model_name == "resnet34":
model = ResNet([3, 4, 6, 3], num_classes=10)
elif model_name == "resnet34_100":
model = ResNet([3, 4, 6, 3], num_classes=100)
elif model_name == "resnet18":
model = models.resnet18()
elif model_name == "resnet50":
model = models.resnet50()
elif model_name == "resnet101":
model = models.resnet101()
elif model_name == "resnet152":
model = models.resnet152()
elif model_name == "wide_resnet101":
model = models.wide_resnet101_2()
elif model_name == "densenet121":
model = DenseNet121(num_classes=10)
elif model_name == "densenet121_100":
model = DenseNet121(num_classes=100)
elif model_name == "matrix_fac_1":
model = LinearNetwork(6, [1], 10, bias=False)
elif model_name == "matrix_fac_4":
model = LinearNetwork(6, [4], 10, bias=False)
elif model_name == "matrix_fac_10":
model = LinearNetwork(6, [10], 10, bias=False)
elif model_name == "linear_fac":
model = LinearNetwork(6, [], 10, bias=False)
return model
def __init__(self,
in_channels,
pretrained=False,
version='resnet101',
clf=None):
super(ResNet_BB, self).__init__()
version = version.strip()
if version == 'resnet18':
resnet = models.resnet18(pretrained)
elif version == 'resnet34':
resnet = models.resnet34(pretrained)
elif version == 'resnet50':
resnet = models.resnet50(pretrained)
elif version == 'resnet101':
resnet = models.resnet101(pretrained)
elif version == 'resnet152':
resnet = models.resnet152(pretrained)
elif version == 'resnext50_32x4d':
resnet = models.resnext50_32x4d(pretrained)
elif version == 'resnext101_32x8d':
resnet = models.resnext101_32x8d(pretrained)
elif version == 'wide_resnet50_2':
resnet = models.wide_resnet50_2(pretrained)
elif version == 'wide_resnet101_2':
resnet = models.wide_resnet101_2(pretrained)
elif version == 'resnest50':
resnet = resnest50(pretrained)
elif version == 'resnest101':
resnet = resnest101(pretrained)
elif version == 'resnest200':
resnet = resnest200(pretrained)
elif version == 'resnest269':
resnet = resnest269(pretrained)
else:
raise NotImplementedError(
'version {} is not supported as of now'.format(version))
resnet.conv1 = nn.Conv2d(in_channels,
64,
kernel_size=(7, 7),
stride=(2, 2),
padding=(3, 3),
bias=False)
if clf == 'deeplabv3Plus':
return_layers = {'layer4': 'out', 'layer1': 'low_level'}
elif clf == 'PointRend':
return_layers = {'layer4': 'out', 'layer2': 'fine_grained'}
else:
return_layers = {'layer4': 'out'}
self.backbone = IntermediateLayerGetter(resnet, return_layers)
self.out_channels = 2048
self.low_channels = 256
self.fine_grained_channels = 512
def get_Oracle(args, chk=False):
model = models.wide_resnet101_2(pretrained=True)
model = model.to(args.device)
model.eval()
if chk is True:
_, test_loader = ImageNet_dataloader(args)
best_acc = test(args, model, args.device, test_loader)
print("\nOracle Acc for all classes=%.2f%%" % (best_acc * 100))
return model
def __init__(self, embed_size, use_images):
super(Encoder, self).__init__()
self.use_images = use_images
resnet = models.wide_resnet101_2(pretrained=True)
self.connection_layer = torch.nn.Sequential(
torch.nn.Linear(resnet.fc.in_features, embed_size),
torch.nn.BatchNorm1d(embed_size),
torch.nn.Dropout(0.5)
)
self.connection_layer.apply(init_weights())
self.resnet = torch.nn.Sequential(*list(resnet.children())[:-1])
def load_wide_resnet(num_classes, device):
model = models.wide_resnet101_2(pretrained=True).to(device)
trained_layer_indices = [7, 9]
for i, child in enumerate(model.children()):
if i not in trained_layer_indices:
for param in child.parameters():
param.requires_grad = False
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, num_classes).to(device)
train_transform, valid_transform = load_transform(image_size=256,
crop_size=224)
return model, train_transform, valid_transform, "pretrained-wide-resnet"
def __init__(self: 'Encoder', z_size: int) -> None:
"""__init__
Initialize the model architecture.
Parameters
----------
z_size: int
size of the latent space ( output size )
"""
super(Encoder, self).__init__()
self.z_size = z_size
self.resnet = models.wide_resnet101_2(pretrained=True)
self._disable_grad()
n_features = self.resnet.fc.in_features
self.resnet.fc = nn.Linear(n_features, z_size)
def load_model(model_name):
if model_name == 'resnet18':
return models.resnet18(pretrained=True)
if model_name == 'resnet34':
return models.resnet34(pretrained=True)
if model_name == 'resnet50':
return models.resnet50(pretrained=True)
if model_name == 'resnet101':
return models.resnet101(pretrained=True)
if model_name == 'resnet152':
return models.resnet152(pretrained=True)
if model_name == 'resnext50_32x4d':
return models.resnext50_32x4d(pretrained=True)
if model_name == 'resnext101_32x8d':
return models.resnext101_32x8d(pretrained=True)
if model_name == 'wide_resnet50_2':
return models.wide_resnet50_2(pretrained=True)
if model_name == 'wide_resnet101_2':
return models.wide_resnet101_2(pretrained=True)
if model_name == 'resnet18_swsl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnet18_swsl')
if model_name == 'resnet50_swsl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnet50_swsl')
if model_name == 'resnext50_32x4d_swsl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnext50_32x4d_swsl')
if model_name == 'resnext101_32x4d_swsl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnext101_32x4d_swsl')
if model_name == 'resnext101_32x8d_swsl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnext101_32x8d_swsl')
if model_name == 'resnext101_32x16d_swsl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnext101_32x16d_swsl')
if model_name == 'resnet18_ssl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnet18_ssl')
if model_name == 'resnet50_ssl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnet50_ssl')
if model_name == 'resnext50_32x4d_ssl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnext50_32x4d_ssl')
if model_name == 'resnext101_32x4d_ssl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnext101_32x4d_ssl')
if model_name == 'resnext101_32x8d_ssl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnext101_32x8d_ssl')
if model_name == 'resnext101_32x16d_ssl':
return torch.hub.load('facebookresearch/semi-supervised-ImageNet1K-models', 'resnext101_32x16d_ssl')
return None
def __init__(self, args: Namespace, dim: int = 2048) -> None:
self.data_dir = args.data_dir
self.images_dir = args.images_dir
with open(path.join(args.data_dir, args.captions)) as infile:
self.captions = infile.readlines()
self.embeddings = np.load(path.join(args.data_dir, args.embeddings))
self.k = k
self.metric = metric
if self.metric == -1:
# Cosine similarity
self.index = faiss.IndexFlatIP(dim)
faiss.normalize_L2(self.embeddings)
self.index.add(self.embeddings)
elif self.metric == 1:
# Euclidean distance (no square root)
self.index = faiss.IndexFlatL2(dim)
self.index.add(self.embeddings)
elif self.metric == 23:
# Mahalanobis distance
self.index = faiss.IndexFlatL2(dim)
x_centered = self.embeddings - self.embeddings.mean(0)
self.transform = np.linalg.inv(np.linalg.cholesky(
np.dot(x_centered.T, x_centered) / x_centered.shape[0])).T
self.index.add(
np.dot(self.embeddings, self.transform).astype(np.float32))
elif self.metric == 0:
# Inner project
self.index = faiss.IndexFlatIP(dim)
self.index.add(self.embeddings)
else:
self.index = faiss.IndexFlat(dim, self.metric)
self.index.add(self.embeddings)
self.model = wide_resnet101_2(pretrained=True, progress=True)
self.model.eval() # Don't forget to put model in evaluation mode!
self.model.fc = Identity()
# Use recommended sequence of transforms for ImageNet pretrained models
self.transforms = Compose([Resize(256, interpolation=Image.BICUBIC), # Default is bilinear
CenterCrop(224),
ToTensor(),
Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])])
def process_network(network_type, model_file, num_classes):
if network_type == 'densenet':
model = models.densenet121(num_classes=num_classes)
elif network_type == 'resnet':
model = models.wide_resnet101_2(num_classes=num_classes)
elif network_type == 'shuffle':
model = models.shufflenet_v2_x2_0(num_classes=num_classes)
print(network_type)
model_params = torch.load(model_file, map_location='cpu')
model.load_state_dict(model_params)
model.eval()
model.cpu()
s = nn.Softmax(dim=-1)
class_pred_all = None
conf_all = None
all_raw_preds = None
with torch.no_grad():
start = time.time()
for i, imgs in enumerate(img_loader):
if i % 2 == 0:
print(f"{i}/{len(img_loader)}")
pred = model(imgs.cpu())
pred = s(pred)
confidence, class_pred = torch.max(pred, 1)
if conf_all is None:
conf_all = confidence
class_pred_all = class_pred
all_raw_preds = pred
else:
conf_all = torch.cat((conf_all, confidence))
class_pred_all = torch.cat((class_pred_all, class_pred))
all_raw_preds = torch.cat((all_raw_preds, pred))
print(time.time() - start)
return conf_all.numpy(), class_pred_all.numpy(), all_raw_preds.numpy()
def load_model(name : str):
if name.startswith('resnext101'):
model = torch.hub.load('pytorch/vision:v0.5.0', 'resnext101_32x8d', pretrained=True)
num_features = model.fc.in_features
model.fc = nn.Linear(num_features, N_CLASSES)
return model
elif name.startswith('resnet152'):
model = models.resnet152(pretrained=True)
num_features = model.fc.in_features
model.fc = nn.Linear(num_features, N_CLASSES)
return model
elif name.startswith('resnet101'):
model = models.resnet101(pretrained=True)
num_features = model.fc.in_features
model.fc = nn.Linear(num_features, N_CLASSES)
return model
elif name.startswith('wide_resnet101'):
model = models.wide_resnet101_2(pretrained=True)
num_features = model.fc.in_features
model.fc = nn.Linear(num_features, N_CLASSES)
return model
elif name.startswith('densenet161'):
model = models.densenet161(pretrained=True)
num_features = model.classifier.in_features
model.classifier = nn.Linear(num_features, N_CLASSES)
return model
elif name.startswith('densenet169'):
model = models.densenet169(pretrained=True)
num_features = model.classifier.in_features
model.classifier = nn.Linear(num_features, N_CLASSES)
return model
elif name.startswith('densenet201'):
model = models.densenet201(pretrained=True)
num_features = model.classifier.in_features
model.classifier = nn.Linear(num_features, N_CLASSES)
return model
else:
raise(ValueError('Select another model'))
def get_pretrained_ResNet(rn_type, classes, mode):
model = None
if rn_type == "18":
model = models.resnet18(pretrained=True)
elif rn_type == "34":
model = models.resnet34(pretrained=True)
elif rn_type == "50":
model = models.resnet50(pretrained=True)
elif rn_type == "101":
model = models.resnet101(pretrained=True)
elif rn_type == "152":
model = models.resnet152(pretrained=True)
elif rn_type == "50Wide":
model = models.wide_resnet50_2(pretrained=True)
elif rn_type == "101Wide":
model = models.wide_resnet101_2(pretrained=True)
if mode not in ["feature", "finetuning"]:
raise RuntimeError("mode must be 'feature' or 'finetuning' ")
if mode == "feature":
for param in model.parameters():
param.requires_grad = False
model.fc = torch.nn.Sequential(
nn.Linear(model.fc.in_features, classes, bias=True), nn.Sigmoid())
return model
def __init__(self, model='ResNet-50'):
super(CNNModel, self).__init__()
if model == 'AlexNet':
print("use AlexNet")
self.features = nn.Sequential(
*list(models.alexnet(pretrained=True).children())[:-2])
elif model == 'ResNet-152':
print("use ResNet-152")
self.features = nn.Sequential(
*list(models.resnet152(pretrained=True).children())[:-2])
elif model == 'ResNeXt-101-32x8d':
print("use ResNetXt-101-32x8d")
self.features = nn.Sequential(
*list(models.resnext101_32x8d(
pretrained=True).children())[:-2])
elif model == 'Wide ResNet-101-2':
print("use Wide ResNet-101-2")
self.features = nn.Sequential(
*list(models.wide_resnet101_2(
pretrained=True).children())[:-2])
else:
print("use default ResNet-50")
self.features = nn.Sequential(
*list(models.resnet50(pretrained=True).children())[:-2])
elif TopModelName=='resnet34':
pre_model = models.resnet34(pretrained=True, progress=True)
elif TopModelName=='resnet50':
pre_model = models.resnet50(pretrained=True, progress=True)
elif TopModelName=='resnet101':
pre_model = models.resnet101(pretrained=True, progress=True)
elif TopModelName=='resnet152':
pre_model = models.resnet152(pretrained=True, progress=True)
elif TopModelName=="resnext50_32x4d":
pre_model = models.resnext50_32x4d(pretrained=True, progress=True)
elif TopModelName=='resnext101_32x8d':
pre_model = models.resnext101_32x8d(pretrained=True, progress=True)
elif TopModelName=="wide_resnet50_2":
pre_model = models.wide_resnet50_2(pretrained=True, progress=True)
elif TopModelName=='wide_resnet101_2':
pre_model = models.wide_resnet101_2(pretrained=True, progress=True)
##############################
elif TopModelName=="mnasnet0_5":
pre_model = models.mnasnet0_5(pretrained=True, progress=True)
elif TopModelName=="mnasnet0_75":
pre_model = models.mnasnet0_75(pretrained=True, progress=True)
elif TopModelName=="mnasnet1_0":
pre_model = models.mnasnet1_0(pretrained=True, progress=True)
elif TopModelName=="mnasnet1_3":
pre_model = models.mnasnet1_3(pretrained=True, progress=True)
elif TopModelName=="efficientnet-b0":
from efficientnet_pytorch import EfficientNet
pre_model = EfficientNet.from_pretrained('efficientnet-b0')
######################### Testing Models #################################
## Test
for model_name in model_name_lst:
# define model:
with torch.no_grad():
if model_name is 'vgg16bn':
model = models.vgg16_bn(pretrained=True)
elif model_name is 'vgg19bn':
model = models.vgg19_bn(pretrained=True)
elif model_name is 'resnet34':
model = models.resnet34(pretrained=True)
elif model_name is 'resnet101':
model = models.resnet101(pretrained=True)
elif model_name is 'resnet152':
model = models.resnet152(pretrained=True)
elif model_name is 'wrn-101-2':
model = models.wide_resnet101_2(pretrained=True)
elif model_name is 'resnext101_32x4d':
model = pretrainedmodels.__dict__[model_name](num_classes=1000, pretrained='imagenet')
elif model_name is 'se_resnet101':
model = pretrainedmodels.__dict__[model_name](num_classes=1000, pretrained='imagenet')
elif model_name is 'senet154':
model = pretrainedmodels.__dict__[model_name](num_classes=1000, pretrained='imagenet')
elif model_name is 'nasnetalarge':
model = pretrainedmodels.__dict__[model_name](num_classes=1000, pretrained='imagenet')
elif model_name is 'pnasnet5large':
model = pretrainedmodels.__dict__[model_name](num_classes=1000, pretrained='imagenet')
elif model_name is 'resnext101_32x48d_wsl':
model = torch.hub.load('facebookresearch/WSL-Images', 'resnext101_32x48d_wsl')
elif model_name is 'effnetE7':
from efficientnet_pytorch import EfficientNet
model = EfficientNet.from_pretrained('efficientnet-b7')
def __init__(self, out_size, pretrained=True):
super().__init__()
model = models.wide_resnet101_2(pretrained=pretrained)
model = list(model.children())[:-1]
model.append(nn.Conv2d(2048, out_size, 1))
self.net = nn.Sequential(*model)
