def validation(epoch):
model.eval()
validation_loss = 0
correct = 0
for data, target in val_loader:
if use_cuda:
data, target = data.cuda(), target.cuda()
output = model(data)
# sum up batch loss
criterion = torch.nn.CrossEntropyLoss(reduction='elementwise_mean')
validation_loss += criterion(output, target).data.item()
# get the index of the max log-probability
pred = output.data.max(1, keepdim=True)[1]
valid = pred.eq(target.data.view_as(pred)).cpu()
wrong_images = np.argwhere(valid.numpy() == 0)[:, 0]
correct += valid.sum()
if epoch > 15:
show_images(data, 5, indexes=wrong_images)
plt.show()
validation_loss /= len(val_loader.dataset)
print(
'\nValidation set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.
format(validation_loss, correct, len(val_loader.dataset),
100. * correct / len(val_loader.dataset)))
def show_move(self, direction, show_full_grid=False):
if isinstance(direction, str):
key_map = {
"W": 1,
"A": 2,
"S": 3,
"D": 0,
"w": 1,
"a": 2,
"s": 3,
"d": 0
}
direction = key_map[direction]
plt.close()
(reward, terminal) = self.move(direction)
# self.get_view()
if show_full_grid:
view = self.view.clone()
grid = self.environment.grid.clone()
view[self.range, self.range] = 1.
grid[tuple(self.environment.agent_point)] = 1.
tools.show_images([view.numpy(), grid.numpy()],
titles=["Agent perspective", "Maze map"])
else:
self.show_view()
return reward, terminal
def unprocess(image, mean_pixel):
return image + mean_pixel
if __name__ == '__main__':
pretrained_vgg_19_path = '/home/meizu/WORK/code/neural_style/model/imagenet-vgg-verydeep-19.mat'
test_image_path = '/home/meizu/WORK/code/neural_style/images/1-content.jpg'
figures_save_path = '/home/meizu/WORK/code/neural_style/figures'
# read image
test_image = tools.imread(test_image_path)
# load pretrained vgg19 weights
weights, mean_pixel = load_weights(pretrained_vgg_19_path)
# preprocess the test image
test_image_preprocessed = np.array([preprocess(test_image, mean_pixel)
]).astype(np.float32)
# infer the input image with vgg19
imput_image_infered = net_infer(weights, test_image_preprocessed)
sess = tf.InteractiveSession()
aa = sess.run(imput_image_infered['relu4_2'])
aa = aa.reshape(aa.shape[1], aa.shape[2], aa.shape[3])
aa_for_plot = []
for i in range(aa.shape[2]):
aa_for_plot.append(aa[:, :, i])
tools.show_images(aa_for_plot[::8], cols=8)
print ''
w, h = gray.shape
rgb = np.empty((w, h, 3), dtype=np.float32)
rgb[:, :, 2] = rgb[:, :, 1] = rgb[:, :, 0] = gray
return rgb
if __name__ == '__main__':
# prepare the content image for style transform
# and the specified style image
content_image_path = '../images/1-content.jpg'
style_image_path = '../images/1-style.jpg'
content_image = tools.imread(content_image_path)
style_image = tools.imread(style_image_path)
style_images = [style_image] # for multi-styles
tools.show_images([content_image, style_image],
titles=['content image', 'style'],
axis='on')
# pretrained vgg path
pre_trained_vgg_path = '../model/imagenet-vgg-verydeep-19.mat'
# set content image width for model input
# and style_scale for scaling the style image
width = 384
if width is not None:
new_shape = (int(
math.floor(
float(content_image.shape[0]) / content_image.shape[1] *
width)), width)
content_image = misc.imresize(content_image,
new_shape).astype(np.float32)
generated_images - content_images) * 2 / tf.to_float(
size) # remain the same as in the paper
return content_loss
def total_variation_loss(layer):
shape = tf.shape(layer)
height = shape[1]
width = shape[2]
y = tf.slice(layer, [0, 0, 0, 0],
tf.stack([-1, height - 1, -1, -1])) - tf.slice(
layer, [0, 1, 0, 0], [-1, -1, -1, -1])
x = tf.slice(layer, [0, 0, 0, 0],
tf.stack([-1, -1, width - 1, -1])) - tf.slice(
layer, [0, 0, 1, 0], [-1, -1, -1, -1])
loss = tf.nn.l2_loss(x) / tf.to_float(
tf.size(x)) + tf.nn.l2_loss(y) / tf.to_float(tf.size(y))
return loss
if __name__ == '__main__':
import argparse
import tools
parser = argparse.ArgumentParser()
parser.add_argument('--conf', default='../conf/candy.yml')
parser.add_argument('--loss_model_file', default='../models/vgg_16.ckpt')
args = parser.parse_args()
FLAGS = utils.read_conf_file(args.conf)
features = get_style_features(FLAGS)
tools.show_images(features, cols=2)
print("")