def validate(self):
self.discriminator.eval()
prediction_image = torch.zeros([self.valloader.dataset.label.shape[0], self.config.patch_shape[0],\
self.config.patch_shape[1], self.config.patch_shape[2]])
whole_vol = self.valloader.dataset.whole_vol
for batch_number, (patches, label,
_) in enumerate(self.valloader.loader):
patches = patches.cuda()
_, batch_prediction_softmax = self.discriminator(patches)
batch_prediction = torch.argmax(batch_prediction_softmax,
dim=1).cpu()
prediction_image[
batch_number * self.config.batch_size:(batch_number + 1) *
self.config.batch_size, :, :, :] = batch_prediction
print("Validating.. [{0}/{1}]".format(
batch_number, self.valloader.num_iterations))
vol_shape_x, vol_shape_y, vol_shape_z = self.config.volume_shape
prediction_image = prediction_image.numpy()
val_image_pred = recompose3D_overlap(prediction_image, vol_shape_x,
vol_shape_y, vol_shape_z,
self.config.extraction_step[0],
self.config.extraction_step[1],
self.config.extraction_step[2])
val_image_pred = val_image_pred.astype('uint8')
pred2d = np.reshape(val_image_pred,
(val_image_pred.shape[0] * vol_shape_x *
vol_shape_y * vol_shape_z))
lab2d = np.reshape(
whole_vol,
(whole_vol.shape[0] * vol_shape_x * vol_shape_y * vol_shape_z))
classes = list(range(0, self.config.num_classes))
F1_score = f1_score(lab2d, pred2d, classes, average=None)
print("Validation Dice Coefficient.... ")
print("Background:", F1_score[0])
print("CSF:", F1_score[1])
print("GM:", F1_score[2])
print("WM:", F1_score[3])
current_validation_dice = F1_score[2] + F1_score[3]
if (self.best_validation_dice < current_validation_dice):
self.best_validation_dice = current_validation_dice
self.save_checkpoint(is_best=True)
def test(self):
self.net.eval()
prediction_image = torch.zeros([self.testloader.dataset.patches.shape[0], self.config.patch_shape[0],\
self.config.patch_shape[1], self.config.patch_shape[2]])
whole_vol = self.testloader.dataset.whole_vol
for batch_number, (patches, _) in enumerate(self.testloader.loader):
patches = patches.cuda()
_, batch_prediction_softmax = self.net(patches)
batch_prediction = torch.argmax(batch_prediction_softmax,
dim=1).cpu()
prediction_image[
batch_number * self.config.batch_size:(batch_number + 1) *
self.config.batch_size, :, :, :] = batch_prediction
print("Testing.. [{0}/{1}]".format(batch_number,
self.testloader.num_iterations))
vol_shape_x, vol_shape_y, vol_shape_z = self.config.volume_shape
prediction_image = prediction_image.numpy()
test_image_pred = recompose3D_overlap(prediction_image, vol_shape_x,
vol_shape_y, vol_shape_z,
self.config.extraction_step[0],
self.config.extraction_step[1],
self.config.extraction_step[2])
test_image_pred = test_image_pred.astype('uint8')
pred2d = np.reshape(test_image_pred,
(test_image_pred.shape[0] * vol_shape_x *
vol_shape_y * vol_shape_z))
lab2d = np.reshape(
whole_vol,
(whole_vol.shape[0] * vol_shape_x * vol_shape_y * vol_shape_z))
classes = list(range(0, self.config.num_classes))
F1_score = f1_score(lab2d, pred2d, classes, average=None)
print("Test Dice Coefficient.... ")
print("Background:", F1_score[0])
print("CSF:", F1_score[1])
print("GM:", F1_score[2])
print("WM:", F1_score[3])