def __init__(self, num_classes=10):
super(LwfModel, self).__init__()
self.num_classes = num_classes
self.known_classes = 0
self.old_net = None
self.net = resnet32(num_classes=num_classes)
self.bce_loss = nn.BCEWithLogitsLoss(reduction='mean')
def get_resnet(resnet=32):
if resnet == 20:
return resnet20()
elif resnet == 32:
return resnet32()
elif resnet == 56:
return resnet56()
else:
raise ValueError("resnet parameter must be 20 32 or 56")
def __init__(self, num_classes=100, memory=2000):
super(iCaRLModel, self).__init__()
self.num_classes = num_classes
self.memory = memory
self.known_classes = 0
self.old_net = None
self.net = resnet32(num_classes=num_classes)
self.bce_loss = nn.BCEWithLogitsLoss(reduction='mean')
self.exemplar_sets = [{
'indexes': [],
'features': []
} for label in range(num_classes)]
self.compute_means = True
self.means = []
def __init__(self,
train_dataset: Cifar100,
num_classes=100,
memory=2000,
batch_size=128,
device='cuda'):
super(iCaRLModel, self).__init__()
self.num_classes = num_classes
self.memory = memory
self.known_classes = 0
self.old_net = None
self.batch_size = batch_size
self.device = device
self.net = resnet32(num_classes=num_classes)
self.dataset = train_dataset
self.bce_loss = nn.BCEWithLogitsLoss(reduction='mean')
self.exemplar_sets = []
self.compute_means = True
self.exemplar_means = []