When I create a fmodel (foolbox model) from a custom pytorch architecture, is there a way to set the default loss of the fmodel to MeanSquaredError?
In other words, is there a parameter for the loss function I can set when defining a fmodel so that whenever a built-in foolbox function (such as accuracy(fmodel, images, labels) or attack(fmodel, images, labels, epsilons=epsilons) ) is called, the function will automatically use the explicitly-defined loss?
Example of fmodel:
#Defining my model
class CNN(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, 3)
def forward(self, x):
x = self.pool(F.relu(self.conv1(x)))
x = self.pool(F.relu(self.conv2(x)))
x = torch.flatten(x, 1)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
x = torch.mean(x)
return x
model = CNN().eval()
# CREATING FMODEL
preprocessing = dict(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5], axis=-3)
fmodel = PyTorchModel(model, bounds=(-1, 1), preprocessing=preprocessing)
When I create a fmodel (foolbox model) from a custom pytorch architecture, is there a way to set the default loss of the fmodel to MeanSquaredError?
In other words, is there a parameter for the loss function I can set when defining a fmodel so that whenever a built-in foolbox function (such as
accuracy(fmodel, images, labels)
orattack(fmodel, images, labels, epsilons=epsilons)
) is called, the function will automatically use the explicitly-defined loss?Example of fmodel: