def __init__(self, n_epochs, lr):
self.n_epochs = n_epochs
self.lr = lr
self.datasets, self.dataloaders = get_ds_loaders()
self.model = CustomNet().to(device)
self.criterion = torch.nn.CrossEntropyLoss()
self.build_opt()
self.vis = Visualizer()
def __init__(self, lr, n_epochs):
self.criterion = nn.CrossEntropyLoss()
self.n_epochs = n_epochs
self.lr = lr
self.build_net()
self.build_opt()
self.datasets, self.dataloaders = get_ds_loaders()
self.vis = Visualizer()
# Device configuration
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
resnet_origin = resnet18(pretrained=True) # original ResNet
without_fc = nn.Sequential(*list(resnet_origin.children())[:-1]).to(
device) # removing last FC layer from ResNet
resnet_ft = torch.load(
"../01_modify_last_FC_layer/ResNet18_CIFAR10_finetuned.pth",
map_location='cpu') # finetuning last FC layer with 10 classes
# print(resnet_origin)
# print(without_fc)
# print(resnet_ft)
datasets, dataloaders = get_ds_loaders()
for images, labels in dataloaders['train']:
images = images.to(device)
labels = labels.to(device)
out1 = resnet_origin(images)
print("ResNet original", out1.shape) # 100, 1000
out2 = without_fc(images)
print("Without FC layer", out2.shape) # 100, 512, 1, 1
out3 = resnet_ft(images)
print("Finetuned FC layer", out3.shape) # 100, 10
break
def __init__(self, ckpt_path):
self.ckpt_path = ckpt_path
self.datasets, self.dataloaders = get_ds_loaders()
self.build_net()
self.load_trained_model()