def compress(autoencoder, image_path, compressed_path):
"""
Represent each of the images as the activities of the narrowest layer
in the autoencoder.
autoencoder: the neural network trained on the images
image_path, compressed_path: str, the directory containing the raw
uncompressed images, and the directory that will hold the
compressed images, respectively
"""
images, imagenames = ldr.load_images(patch_size, image_path)
for i_image, image in enumerate(images):
compressed_filename = os.path.join(compressed_path,
imagenames[i_image] + ".pkl")
compressed_image = None
n_rows, n_cols = image.shape
n_patch_rows = int(n_rows / patch_size)
n_patch_cols = int(n_cols / patch_size)
for i_row in np.arange(n_patch_rows):
for i_col in np.arange(n_patch_cols):
patch = image[i_row * patch_size:(i_row + 1) * patch_size,
i_col * patch_size:(i_col + 1) * patch_size]
compressed_patch = autoencoder.forward_pass(
patch,
evaluating=True,
i_stop_layer=3,
)
if compressed_image is None:
compressed_image = np.zeros(
(n_patch_rows, n_patch_cols, compressed_patch.size))
compressed_image[i_row, i_col, :] = compressed_patch
with open(compressed_filename, "wb") as f:
pkl.dump(compressed_image, f)
# Forward pass
test_outputs = net(inputs)
test_loss_size = loss(test_outputs, labels.long())
total_test_loss += test_loss_size.data.item()
test_loss_hist.append(total_test_loss / len(test_loader))
test_hist_x.append(idx)
print("Validation loss = {:.2f}".format(total_test_loss /
len(test_loader)))
print("Training finished, took {:.2f}s".format(time.time() -
training_start_time))
return train_hist_x, train_loss_hist, test_hist_x, test_loss_hist
if __name__ == "__main__":
dataset = image_loader.load_images("dataset1", (image_h, image_w))
dataset_train, dataset_test = sklearn.model_selection.train_test_split(
dataset, test_size=0.25)
train_loader = make_loader(dataset_train, 32)
test_loader = make_loader(dataset_test, 32)
device = "cpu" # Change to "cuda" if on a machine with cuda
net = cnn()
net.to(device)
train_hist_x, train_loss_hist, test_hist_x, test_loss_hist = train_model(
net, 10, 1e-2, train_loader, test_loader)
"""
Created on Sun Mar 26 14:13:45 2017
@author: phuem
"""
#http://scikit-learn.org/stable/modules/svm.html
from sklearn import svm, metrics
import image_loader
import numpy as np
import matplotlib.pyplot as plt
#------------------------------------------------------------------------------
#Load and reshape Images with image_loader.py
training_set, training_labels = image_loader.load_images("images/train/")
test_set, test_labels = image_loader.load_images("images/test/")
resize_set, resize_labels = image_loader.prep_datas(training_set,
training_labels)
resize_test_set, resize_test_labels = image_loader.prep_datas(
test_set, test_labels)
#---------------------------------------------------------------
#very ugly way to bring vectors to the right shape for SVC fit()
a = []
for x in resize_set:
a.append(x.tolist())
#----------------------------------------------------------------
X = a #reshaped images (training)
y = resize_labels #labels
import sys
import tensorflow as tf
import numpy as np
from matplotlib import patches as patches
from matplotlib import pyplot as plt
from alexnet import AlexNet
from image_loader import load_images
files = sys.argv[1:]
imgs = load_images(files)
m, h, w, _ = imgs.shape
# load the saved tensorflow model and evaluate a list of paths to PNG files (must be 150x150)
num_classes = 1
X = tf.placeholder(tf.float32, shape=(m, h, w, 1))
Y = tf.placeholder(tf.float32, shape=(m, num_classes))
dropout = tf.placeholder(tf.float32)
model = AlexNet(X, dropout, num_classes)
predictions = model.logits > 0
saver = tf.train.Saver()
with tf.Session() as sess:
saver.restore(sess, "./tensorflow-ckpt/model.ckpt")
logits = sess.run(model.logits, feed_dict={X: imgs, dropout: 0})
_, y_h, y_w, __ = logits.shape
import sys
import tensorflow as tf
from alexnet import AlexNet
from image_loader import load_images
# load the saved tensorflow model and evaluate a list of paths to PNG files (must be 150x150)
num_classes = 1
X = tf.placeholder(tf.float32, shape=(None, 150, 150, 1))
Y = tf.placeholder(tf.float32, shape=(None, num_classes))
dropout = tf.placeholder(tf.float32)
model = AlexNet(X, dropout, num_classes)
predictions = model.logits > 0
saver = tf.train.Saver()
files = sys.argv[1:]
imgs = load_images(files, resize=True, remove_alpha=True)
with tf.Session() as sess:
saver.restore(sess, "./tensorflow-ckpt/model.ckpt")
preds = sess.run(predictions, feed_dict={X: imgs, dropout: 0})
print([
str(i) + ": " + ("molecule" if pred else "non-molecule")
for i, pred in enumerate(preds)
])
X = tf.placeholder(tf.float32, shape=(None, 150, 150, 1))
dropout = tf.constant(0)
model = AlexNet(X, dropout, num_classes)
graph = tf.get_default_graph()
# from our graph, get the tensors for our hidden layers
first_hidden_layer_tensor = graph.get_tensor_by_name("conv1/Relu:0")
second_hidden_layer_tensor = graph.get_tensor_by_name("conv2/Relu:0")
third_hidden_layer_tensor = graph.get_tensor_by_name("conv3/Relu:0")
saver = tf.train.Saver()
imgs = load_images([
"./positives/images/*.png",
"./positives/manually-collected/*.png"
])
with tf.Session() as sess:
saver.restore(sess, "./tensorflow-ckpt/model.ckpt")
first_hidden_layer, second_hidden_layer, third_hidden_layer = sess.run((first_hidden_layer_tensor,
second_hidden_layer_tensor,
third_hidden_layer_tensor),
feed_dict={X: imgs})
def deconvolve_layer_1(rng, ignore_pad=False):
return deconvolve_range(rng, filt_size=7, stride=4)
def deconvolve_layer_2(rng, ignore_pad=False):
patch_range = deconvolve_range(rng, filt_size=5, stride=1, pad=(0 if ignore_pad else 2))
def run(program_root):
lgr.debug("Script start")
with ConfigEnvironment(conf):
images = load_images()
if images is None or len(images) == 0:
return
# Check if the images are valid
res = reduce(lambda a, b: a and b.is_valid, images, True)
if not res:
lgr.debug("Find invalid image name.")
show_message_box(
"发现不正确的图片名称，注意图片名称应当符合“名字_装裱_作者_朝代.jpg”（或png)的格式，且请注意"
"附上封面，封面的名称应该为“名字_装裱_作者_朝代_cover.jpg”（或者png）的格式。")
return
if not need_recreate_motion_xml(images):
lgr.debug("No update is found.")
return
else:
daily_image_dir = \
prepare_image_directories(program_root)
if daily_image_dir is None:
lgr.debug("Fail to create destination folders")
show_message_box("无法创建目标文件夹")
return
# copy images into this dir
lgr.debug("Copying images into destination folder")
for image in images:
src_abs_path = image.image_path
dst_abs_path = os.path.join(daily_image_dir, image.image_name)
lgr.debug("Copying %s" % image.image_name)
shutil.copy(src_abs_path, dst_abs_path)
# copy cover if exits
src_cover_abs_path = image.get_cover_path()
if os.path.exists(src_cover_abs_path):
dst_cover_abs_path = os.path.join(
daily_image_dir,
os.path.basename(image.get_cover_path()))
lgr.debug("Find cover and copying %s" %
os.path.basename(image.get_cover_path()))
shutil.copy(src_cover_abs_path, dst_cover_abs_path)
image.move_to(daily_image_dir)
# generate thumbnail
lgr.debug("Creating thumbnail")
build_thumbnail(image)
lgr.debug("Done copying.")
lgr.debug("Creating motion.xml file")
# create motion file
create_motion_xml(images, program_root)
# copy motion file into conf folder of the WPF program
lgr.debug("Moving motion.xml to WPF program folder")
conf_dir = os.path.join(program_root, "conf")
dst_motion_path = os.path.join(conf_dir, "motion.xml")
if os.path.exists(dst_motion_path):
os.remove(dst_motion_path)
shutil.copy(get_motion_xml_path(program_root), dst_motion_path)
# update conf and save history
conf.update_history(map(lambda aw: aw.artwork_full_name, images))
lgr.debug("Done!\n\n\n")
show_message_box("完成！")