def show_images(images, title=None, path=None):
images = utils.make_grid(images)
show_image(images[0], title, path)
reconstruction = model(images)
# print(test_set_reconstruction.shape)
recon_image_ = reconstruction.view(reconstruction.shape[0], 1, 28,
28)
images = images.view(images.shape[0], 1, 28, 28)
if r % 100 == 0:
# show_images(images, 'original')
# show_images(recon_image_, 'test_set_reconstruction')
grid1 = torchvision.utils.make_grid(images)
writer.add_image('orig images', grid1, 0)
grid2 = torchvision.utils.make_grid(recon_image_)
writer.add_image('recon images', grid2)
writer.close()
## maybe we just store the test_set_reconstruction
## maybe we just store the test_set_reconstruction
images = utils.make_grid(images)
recon_image_ = utils.make_grid(recon_image_)
plt.imshow(images[0], cmap='gray')
plt.title('Original from epoch {}'.format(epoch + 1))
plt.savefig(
'reconstruction_during_training/originals_epoch_{}_example_{}'
.format(epoch + 1, r))
plt.imshow(recon_image_[0], cmap='gray')
plt.title('Reconstruction from epoch {}'.format(epoch + 1))
plt.savefig(
'reconstruction_during_training/reconstruction_epoch_{}_example_{}'
.format(epoch + 1, r))
model.eval()
## we want also to sample something from the model during training
rendom_samples = model.sample(N_SAMPLE)
test_loss_all, test_loss_localization, test_loss_cover = \
localization_loss(pred_label, cropout_label, test_hidden, test_cover, beta=1)
# diff_S, diff_C = np.abs(np.array(test_output.data[0]) - np.array(test_secret.data[0])), np.abs(np.array(test_hidden.data[0]) - np.array(test_cover.data[0]))
# print (diff_S, diff_C)
if idx in [1, 2, 3, 4]:
print('Training: Batch {0}/{1}. Total Loss {2:.4f}, Localization Loss {3:.4f}, Cover Loss {3:.4f} '.format(
idx + 1, len(train_loader), test_loss_all.data, test_loss_localization.data, test_loss_cover.data))
# Creates img tensor
# imgs = [test_secret.data, test_cover.data, test_hidden.data, test_output.data] # 隐藏图像 宿主图像 输出图像 提取得到的图像
imgs = [test_cover.data, test_hidden.data]
imgs_tsor = torch.cat(imgs, 0)
# Prints Images
imshow(utils.make_grid(imgs_tsor), idx + 1, learning_rate=learning_rate, beta=beta)
# target_tensor = torch.tensor((pred_label.reshape(1,14,14).detach().cpu().numpy()*255).astype(np.uint8)).to(device)
# imshow(target_tensor, idx+1, learning_rate=learning_rate, beta=beta)
print('Expected:')
print(pred_label.data)
print('Real:')
print(cropout_label.data)
test_losses.append(test_loss_all.data.cpu().numpy())
mean_test_loss = np.mean(test_losses)
print('Average loss on test set: {:.2f}'.format(mean_test_loss))
test_hidden, pred_label, cropout_label = net(test_secret, test_cover)
# MSE标签距离 loss
test_loss_all, test_loss_localization, test_loss_cover = \
localization_loss(pred_label, cropout_label, test_hidden, test_cover, beta=1)
# diff_S, diff_C = np.abs(np.array(test_output.data[0]) - np.array(test_secret.data[0])), np.abs(np.array(test_hidden.data[0]) - np.array(test_cover.data[0]))
# print (diff_S, diff_C)
if idx in [1, 2, 3, 4]:
print(
'Training: Batch {0}/{1}. Total Loss {2:.4f}, Localization Loss {3:.4f}, Cover Loss {3:.4f} '
.format(idx + 1, len(train_loader), test_loss_all.data,
test_loss_localization.data, test_loss_cover.data))
# Creates img tensor
# imgs = [test_secret.data, test_cover.data, test_hidden.data, test_output.data] # 隐藏图像 宿主图像 输出图像 提取得到的图像
imgs = [test_secret, test_hidden, pred_label]
imgs_tsor = torch.cat(imgs, 0)
# Prints Images
imshow(utils.make_grid(imgs_tsor),
idx + 1,
learning_rate=learning_rate,
beta=beta)
test_losses.append(test_loss_all.data.cpu().numpy())
mean_test_loss = np.mean(test_losses)
print('Average loss on test set: {:.2f}'.format(mean_test_loss))
