def __init__(self, sess, img_h, img_w, img_c, op):
# input
self.sess = sess
self.output_height, self.output_width = img_h, img_w
self.c_dim = img_c
self.orig_embed_size = 4800
# input batch
self.op = op
if self.op == "train":
self.batch_size = 64
print "loading training data......"
with open("./train_data/img_objs.pk", "r") as f:
img_objs = pk.load(f)
self.data_size = len(img_objs)
print "number of image {}".format(self.data_size)
self.batch_num = self.data_size / self.batch_size
print "number of batch {}".format(self.batch_num)
batch = data_reader.get_train_batch(img_objs, self.batch_size)
self.img_batch = batch[0]
self.wimg_batch = batch[1]
self.match_embed_batch = batch[2]
self.mismatch_embed_batch = batch[3]
elif self.op == "test":
self.batch_size = 1
print "loading testing data"
self.test_sent = tf.placeholder(
tf.float32, shape=[self.batch_size, self.orig_embed_size])
# network setting
self.gf_dim = 64
self.df_dim = 64
self.z_dim = 100
self.embed_size = 128
self.keep_prob = tf.placeholder(tf.float32)
# batch_norm of discriminator
self.d_bn0 = batch_norm(name="d_bn0")
self.d_bn1 = batch_norm(name="d_bn1")
self.d_bn2 = batch_norm(name="d_bn2")
self.d_bn3 = batch_norm(name="d_bn3")
self.d_bn4 = batch_norm(name="d_bn4")
# batch_norm of generator
self.g_bn0 = batch_norm(name="g_bn0")
self.g_bn1 = batch_norm(name="g_bn1")
self.g_bn2 = batch_norm(name="g_bn2")
self.g_bn3 = batch_norm(name="g_bn3")
# build model
print "building model......"
self.build_model()
def __init__(self, sess, img_h, img_w, img_c, op):
#---input setting---#
self.sess = sess
self.op = op
self.output_height, self.output_width = img_h, img_w
self.c_dim = img_c
self.orig_embed_size = 4800
#---training data---#
if op == "train":
self.batch_size = 64
print "loading training data......"
with open("./train_data/img_objs_new.pk", "r") as f:
img_objs = pk.load(f)
batch = data_reader.get_train_batch(img_objs, self.batch_size)
self.rimg_batch = batch[0]
self.wimg_batch = batch[1]
self.match_embed_batch = batch[2]
self.mismatch_embed_batch = batch[3]
#---testing data---#
if op == "test":
self.batch_size = 1
self.test_sent = tf.placeholder(tf.float32, shape=
[1, self.orig_embed_size])
#---model network setting---#
self.gf_dim = 64
self.df_dim = 64
self.z_dim = 100
self.embed_size = 128
self.keep_prob = tf.placeholder(tf.float32)
#---batch_norm of discriminator---#
self.d_bn0 = batch_norm(name="d_bn0")
self.d_bn1 = batch_norm(name="d_bn1")
self.d_bn2 = batch_norm(name="d_bn2")
self.d_bn3 = batch_norm(name="d_bn3")
self.d_bn4 = batch_norm(name="d_bn4")
#---batch_norm of generator---#
self.g_bn0 = batch_norm(name="g_bn0")
self.g_bn1 = batch_norm(name="g_bn1")
self.g_bn2 = batch_norm(name="g_bn2")
self.g_bn3 = batch_norm(name="g_bn3")
#---build model---#
print "building model......"
self.build_model()
aim_mod = AIM_gen.AIM_gen()
# get data_reader
import data_reader
data_reader = data_reader.data_reader('outpt.txt')
BSIZE = 32
ITER_PER_EPOC = 200000 // BSIZE
EPOC = 50
MAX_ITER = ITER_PER_EPOC * EPOC
ETA = M.ETA(MAX_ITER)
ETA.start()
for iteration in range(MAX_ITER + 1):
img, age_fake, age = data_reader.get_train_batch(BSIZE)
losses = aim_mod.train(img, age_fake, age, normalize=True)
if iteration % 10 == 0:
print('------ Iteration %d ---------' % iteration)
aim_mod.display_losses(losses)
print('ETA', ETA.get_ETA(iteration))
if iteration % 1000 == 0 and iteration > 0:
aim_mod.save('%d.ckpt' % iteration)
if iteration % 100 == 0:
gen = aim_mod.eval(img, age_fake)
img = np.uint8(img)
gen = np.uint8(gen)
for i in range(BSIZE):
cv2.imwrite('./res/%d_%d_1.jpg' % (iteration, i), img[i])
# get data_reader
import data_reader
data_reader = data_reader.data_reader('outpt.txt')
M.set_gpu('1')
BSIZE = 32
ITER_PER_EPOC = 200000 // BSIZE
EPOC = 30
MAX_ITER = ITER_PER_EPOC * EPOC
aim_mod = AIM.AIM(data_reader.age_class, data_reader.max_id)
ETA = M.ETA(MAX_ITER)
ETA.start()
for iteration in range(MAX_ITER + 1):
img, target, uniform, age, idn = data_reader.get_train_batch(BSIZE)
losses, generated = aim_mod.train(img,
target,
uniform,
age,
idn,
normalize=True)
if iteration % 10 == 0:
print('------ Iteration %d ---------' % iteration)
aim_mod.display_losses(losses)
print('ETA', ETA.get_ETA(iteration))
if iteration % 1000 == 0 and iteration > 0:
aim_mod.save('%d.ckpt' % iteration)