def color(data, template, logs_dir="/data2/ben/HE/data/result/color_logs/"):
import tensorflow.compat.v1 as tf
tf.reset_default_graph()
sess = tf.Session()
image_dist = []
is_train = False
#print("FLAGS",tf.flags.FLAGS)
#print("FLAGS.logs_dir==============",tf.flags.FLAGS.logs_dir)
config = get_config(FLAGS, is_train)
dist = DCGMM(sess, config, "DCGMM", is_train)
db_tmpl = SampleProvider("Template_dataset", template, config.fileformat,
config.image_options, is_train)
mu_tmpl = 0
std_tmpl = 0
N = 0
while True:
X = db_tmpl.DrawSample(config.batch_size)
if len(X) == 0:
break
X_hsd = utils.RGB2HSD(X / 255.0)
mu, std, gamma = dist.deploy(X_hsd)
mu = np.asarray(mu)
mu = np.swapaxes(mu, 1, 2) # -> dim: [ClustrNo x 1 x 3]
std = np.asarray(std)
std = np.swapaxes(std, 1, 2) # -> dim: [ClustrNo x 1 x 3]
N = N + 1
mu_tmpl = (N - 1) / N * mu_tmpl + 1 / N * mu
std_tmpl = (N - 1) / N * std_tmpl + 1 / N * std
db = SampleProvider("Test_dataset", data, config.fileformat,
config.image_options, is_train)
while True:
X = db.DrawSample(config.batch_size)
if len(X) == 0:
break
X_hsd = utils.RGB2HSD(X / 255.0)
mu, std, pi = dist.deploy(X_hsd)
mu = np.asarray(mu)
mu = np.swapaxes(mu, 1, 2) # -> dim: [ClustrNo x 1 x 3]
std = np.asarray(std)
std = np.swapaxes(std, 1, 2) # -> dim: [ClustrNo x 1 x 3]
X_conv = image_dist_transform(X_hsd, mu, std, pi, mu_tmpl, std_tmpl,
config.im_size, config.ClusterNo)
image_dist.append(X_conv)
# misc.imsave('/data2/ben/HE/data/result/color_tmp/color_norma.png', np.squeeze(X_conv))
return image_dist
def main():
sess = tf.Session()
if FLAGS.mode == "train":
is_train = True
else:
is_train = False
config = get_config(FLAGS, is_train)
if not os.path.exists(config.logs_dir):
os.makedirs(config.logs_dir)
dist = DCGMM(sess, config, "DCGMM", is_train)
db = SampleProvider("Train_dataset", config.data_dir, config.fileformat, config.image_options, is_train)
if FLAGS.mode == "train":
for i in range(int(config.iteration)):
X = db.DrawSample(config.batch_size)
X_hsd = utils.RGB2HSD(X[0]/255.0)
loss, summary_str, summary_writer = dist.fit(X_hsd)
if i % config.ReportInterval == 0:
summary_writer.add_summary(summary_str, i)
print("iter {:>6d} : {}".format(i+1, loss))
if i % config.SavingInterval == 0:
dist.saver.save(sess, config.logs_dir+ "model.ckpt", i)
elif FLAGS.mode == "prediction":
if not os.path.exists(config.out_dir):
os.makedirs(config.out_dir)
db_tmpl = SampleProvider("Template_dataset", config.tmpl_dir, config.fileformat, config.image_options, is_train)
mu_tmpl = 0
std_tmpl = 0
N = 0
while True:
X = db_tmpl.DrawSample(config.batch_size)
if len(X) ==0:
break
X_hsd = utils.RGB2HSD(X[0]/255.0)
mu, std, gamma = dist.deploy(X_hsd)
mu = np.asarray(mu)
mu = np.swapaxes(mu,1,2) # -> dim: [ClustrNo x 1 x 3]
std = np.asarray(std)
std = np.swapaxes(std,1,2) # -> dim: [ClustrNo x 1 x 3]
N = N+1
mu_tmpl = (N-1)/N * mu_tmpl + 1/N* mu
std_tmpl = (N-1)/N * std_tmpl + 1/N* std
print("Estimated Mu for template(s):")
print(mu_tmpl)
print("Estimated Sigma for template(s):")
print(std_tmpl)
db = SampleProvider("Test_dataset", config.data_dir, config.fileformat, config.image_options, is_train)
while True:
X = db.DrawSample(config.batch_size)
if len(X) ==0:
break
X_hsd = utils.RGB2HSD(X[0]/255.0)
mu, std, pi = dist.deploy(X_hsd)
mu = np.asarray(mu)
mu = np.swapaxes(mu,1,2) # -> dim: [ClustrNo x 1 x 3]
std = np.asarray(std)
std = np.swapaxes(std,1,2) # -> dim: [ClustrNo x 1 x 3]
X_conv = image_dist_transform(X_hsd, mu, std, pi, mu_tmpl, std_tmpl, config.im_size, config.ClusterNo)
filename = X[1]
filename = filename[0].split('/')[-1]
print(filename)
if not os.path.exists(config.out_dir):
os.makedirs(config.out_dir)
misc.imsave(config.out_dir+filename, np.squeeze(X_conv))
else:
print('Invalid "mode" string!')
return
def main():
sess = tf.Session()
if FLAGS.mode == "train":
is_train = True
else:
is_train = False
config = get_config(FLAGS, is_train)
if not os.path.exists(config.logs_dir):
os.makedirs(config.logs_dir)
dist = DCGMM(sess, config, "DCGMM", is_train)
config.fileformat = "png"
db = SampleProvider("Train_dataset", config.data_dir, config.fileformat, config.image_options, is_train)
if FLAGS.mode == "train":
for i in range(int(config.iteration)):
X = db.DrawSample(config.batch_size)
X_hsd = utils.RGB2HSD(X[0]/255.0)
loss, summary_str, summary_writer = dist.fit(X_hsd)
if i % config.ReportInterval == 0:
summary_writer.add_summary(summary_str, i)
print("iter {:>6d} : {}".format(i+1, loss))
if i % config.SavingInterval == 0:
dist.saver.save(sess, config.logs_dir+ "model.ckpt", i)
elif FLAGS.mode == "prediction":
if not os.path.exists(config.out_dir):
os.makedirs(config.out_dir)
db_tmpl = SampleProvider("Template_dataset", config.tmpl_dir, config.fileformat, config.image_options, is_train)
print(config.fileformat)
mu_tmpl = 0
std_tmpl = 0
N = 0
while True:
X = db_tmpl.DrawSample(config.batch_size)
if len(X) ==0:
break
X_hsd = utils.RGB2HSD(X[0]/255.0)
mu, std, gamma = dist.deploy(X_hsd)
mu = np.asarray(mu)
mu = np.swapaxes(mu,1,2) # -> dim: [ClustrNo x 1 x 3]
std = np.asarray(std)
std = np.swapaxes(std,1,2) # -> dim: [ClustrNo x 1 x 3]
N = N+1
mu_tmpl = (N-1)/N * mu_tmpl + 1/N* mu
std_tmpl = (N-1)/N * std_tmpl + 1/N* std
print("Estimated Mu for template(s):")
print(mu_tmpl)
print("Estimated Sigma for template(s):")
print(std_tmpl)
db = SampleProvider("Test_dataset", config.data_dir, config.fileformat, config.image_options, is_train)
while True:
X = db.DrawSample(config.batch_size)
if len(X) ==0:
break
X_hsd = utils.RGB2HSD(X[0]/255.0)
mu, std, pi = dist.deploy(X_hsd)
mu = np.asarray(mu)
mu = np.swapaxes(mu,1,2) # -> dim: [ClustrNo x 1 x 3]
std = np.asarray(std)
std = np.swapaxes(std,1,2) # -> dim: [ClustrNo x 1 x 3]
X_conv = image_dist_transform(X_hsd, mu, std, pi, mu_tmpl, std_tmpl, config.im_size, config.ClusterNo)
filename = X[1]
filename = filename[0].split('/')[-1]
print(filename)
if not os.path.exists(config.out_dir):
os.makedirs(config.out_dir)
misc.imsave(config.out_dir+filename.replace('datasett\\',''), np.squeeze(X_conv))
else:
print('Invalid "mode" string!')
return
def main():
sess = tf.Session()
image_dist = []
if FLAGS.mode == "train":
is_train = True
else:
is_train = False
config = get_config(FLAGS, is_train)
dist = DCGMM(sess, config, "DCGMM", is_train)
if FLAGS.mode == "train":
db = SampleProvider("Train_dataset", config.data_dir, config.fileformat, config.image_options, is_train)
for i in range(int(config.iteration)):
X = db.DrawSample(config.batch_size)
X_hsd = utils.RGB2HSD(X[0]/255.0)
loss, summary_str, summary_writer = dist.fit(X_hsd)
if i % config.ReportInterval == 0:
summary_writer.add_summary(summary_str, i)
print("iter {:>6d} : {}".format(i+1, loss))
if i % config.SavingInterval == 0:
dist.saver.save(sess, config.logs_dir+ "model.ckpt", i)
elif FLAGS.mode == "prediction":
db_tmpl = SampleProvider("Template_dataset", config.tmpl_dir, config.fileformat, config.image_options, is_train)
print("db_tmpl")
print(db_tmpl)
mu_tmpl = 0
std_tmpl = 0
N = 0
while True:
X = db_tmpl.DrawSample(config.batch_size)
if len(X) == 0:
break
X_hsd = utils.RGB2HSD(X[0]/255.0)
mu, std, gamma = dist.deploy(X_hsd)
mu = np.asarray(mu)
mu = np.swapaxes(mu,1,2) # -> dim: [ClustrNo x 1 x 3]
std = np.asarray(std)
std = np.swapaxes(std,1,2) # -> dim: [ClustrNo x 1 x 3]
N += 1
mu_tmpl = (N-1)/N * mu_tmpl + 1/N* mu
std_tmpl = (N-1)/N * std_tmpl + 1/N* std
print("Estimated Mu for template(s):")
print(mu_tmpl)
print("Estimated Sigma for template(s):")
print(std_tmpl)
image_dir_list = glob(join(config.data_dir, r'*/'))
for image_dir in image_dir_list:
print(image_dir)
png_addr = glob(join(image_dir, '*CN.png'))
if png_addr != []:
continue
db = SampleProvider("Test_dataset", image_dir, config.fileformat, config.image_options, is_train)
while True:
X = db.DrawSample(config.batch_size)
if len(X) ==0:
break
X_hsd = utils.RGB2HSD(X[0]/255.0)
mu, std, pi = dist.deploy(X_hsd)
mu = np.asarray(mu)
mu = np.swapaxes(mu,1,2) # -> dim: [ClustrNo x 1 x 3]
std = np.asarray(std)
std = np.swapaxes(std,1,2) # -> dim: [ClustrNo x 1 x 3]
X_conv = image_dist_transform(X_hsd, mu, std, pi, mu_tmpl, std_tmpl, config.im_size, config.ClusterNo)
image_dist.append(X_conv)
filename = X[1]
misc.imsave(filename[0].split('.orig')[0]+'.CN.png', np.squeeze(X_conv))
print(len(image_dist))
else:
print('Invalid "mode" string!')
return