def __call__(self, tensor):
"""
Args:
tensor (Tensor): Tensor image of size (C, H, W) to be normalized.
Returns:
Tensor: Normalized image.
"""
return F.normalize(tensor, self.mean, self.std)
def forward(self, Z, return_y=False):
expd = self.E(Z)
comp = torch.stack([C(Z) for C in self.Cs])
clus, y_approx = self.nonlinear(comp)
Z = Z + self.eta * (expd - clus)
Z = F.normalize(Z)
if return_y:
return Z, y_approx
return Z
def postprocess(self, X):
return F.normalize(X)
def __call__(self, image, target):
image = F.normalize(image, mean=self.mean, std=self.std)
return image, target
def postprocess(self, X):
Z = ifft2(X, norm='ortho', dim=(2, 3))
return F.normalize(Z).real
def preprocess(self, X):
Z = F.normalize(X)
return fft2(Z, norm='ortho', dim=(2, 3))
batch_size=args.batch_size,
loss=args.loss,
device=device)
X_train, y_train, Z_train = F.to_cpu(X_train, y_train, Z_train)
utils.save_loss(eval_dir, f'train', net.get_loss())
print('test')
Z_test = net.batch_forward(X_test,
batch_size=args.batch_size,
loss=args.loss,
device=device)
X_test, y_test, Z_test = F.to_cpu(X_test, y_test, Z_test)
utils.save_loss(eval_dir, f'test', net.get_loss())
## Normalize
X_train = F.normalize(X_train.flatten(1))
X_test = F.normalize(X_test.flatten(1))
Z_train = F.normalize(Z_train.flatten(1))
Z_test = F.normalize(Z_test.flatten(1))
# Evaluate
evaluate.evaluate(eval_dir, 'knn', Z_train, y_train, Z_test, y_test)
#evaluate.evaluate(eval_dir, 'nearsub', Z_train, y_train, Z_test, y_test, num_classes=num_classes, n_comp=10)
# Plot
plot.plot_loss_mcr(eval_dir, 'train')
plot.plot_loss_mcr(eval_dir, 'test')
plot.plot_heatmap(eval_dir, 'X_train', X_train, y_train, num_classes)
plot.plot_heatmap(eval_dir, 'X_test', X_test, y_test, num_classes)
plot.plot_heatmap(eval_dir, 'Z_train', Z_train, y_train, num_classes)
plot.plot_heatmap(eval_dir, 'Z_test', Z_test, y_test, num_classes)
