def __getitem__(self, idx):
(image, mask) = self.pair[idx]
x_tensor, y_tensor = self.ToTensor(
image, MyFunctions.ImageRescale(mask, [0, 1]))
return x_tensor, y_tensor
def __getitem__(self, idx):
(image, mask) = self.pair[idx]
# only load the original noisy image channel
x_tensor, y_tensor = self.ToTensor(
image[0, :, :], MyFunctions.ImageRescale(mask, [0, 1]))
return x_tensor, y_tensor
volume_sf = np.zeros([486, 1024, 500], dtype=np.float32)
volume_dn = np.zeros([486, 1024, 500], dtype=np.float32)
for step, [img, mask] in enumerate(test_loader):
model.eval()
with torch.no_grad():
x = Variable(img).to(device)
pred_y = model(x)
pred_y = pred_y.detach().cpu().numpy()
x = img.numpy()
y = mask.numpy()
volume_x[step, :, :] = x[0, 0, :, :500]
volume_y[step, :, :] = MyFunctions.ImageRescale(
y[0, 0, :, :500], [0, 255])
volume_sf[step, :, :] = x[0, 2, :, :500]
volume_dn[step, :, :] = MyFunctions.ImageRescale(
pred_y[0, 0, :, :500], [0, 255])
if step % 20 == 0:
plt.figure(figsize=(18, 15))
plt.subplot(1, 4, 1), plt.axis('off'),
plt.imshow(x[0, 0, :, :], cmap='gray'), plt.title('noisy input')
plt.subplot(1, 4, 2), plt.axis('off'),
plt.imshow(x[0, 2, :, :], cmap='gray'), plt.title('self-fusion')
plt.subplot(1, 4, 4), plt.axis('off'),
plt.imshow(y[0, 0, :, :], cmap='gray'), plt.title('5-average')
plt.subplot(1, 4, 3), plt.axis('off'),
plt.imshow(pred_y[0, 0, :, :], cmap='gray'), plt.title('denoise')
plt.show()
G_error.item(), D_x, D_G_z1, D_G_z2))
print(
'BCE_loss: %.4f\tL1_loss: %.4f\tL2_loss: %.4f' %
(BCE_loss(output, label).item(), alpha *
L1_loss(fake_y, y).item(), beta * MSE_loss(fake_y, y).item()))
netG.eval()
if step == 0:
with torch.no_grad():
# Train data denoising
denoise_train = netG(x).detach().cpu().numpy()
noise_train = train_x.numpy()
avg_train = train_y.numpy()
img_x = MyFunctions.ImageRescale(noise_train[0, 0, :, :500],
[0, 255])
img_y = MyFunctions.ImageRescale(avg_train[0, 0, :, :500],
[0, 255])
img_sf = MyFunctions.ImageRescale(noise_train[0, 2, :, :500],
[0, 255])
img_dn = MyFunctions.ImageRescale(denoise_train[0, 0, :, :500],
[0, 255])
plt.figure(figsize=(18, 15))
plt.subplot(1, 4, 1), plt.axis('off'),
plt.imshow(img_x, cmap='gray'), plt.title('noisy')
plt.subplot(1, 4, 2), plt.axis('off'),
plt.imshow(img_sf, cmap='gray'), plt.title('self-fused')
plt.subplot(1, 4, 3), plt.axis('off'),
plt.imshow(img_dn, cmap='gray'), plt.title('denoised')
plt.subplot(1, 4, 4), plt.axis('off'),
