def main():
config = Config()
with tf.Graph().as_default():
# create a reader object
reader = read_data.PoseReader("./labels/txt/new_annos.txt",
"./data/test_imgs/", config)
# create a model object
model = cpm.CPM(config)
# feedforward
predicts = model.build_fc(False)
# return the loss
loss = model.loss()
# Initializing operation
init_op = tf.global_variables_initializer()
saver = tf.train.Saver(max_to_keep = 100)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
sess.run(init_op)
model.restore(sess, saver, config.load_filename)
# start testing
for idx in xrange(config.test_num):
with tf.device("/cpu:0"):
imgs, fm, coords, begins, filename_list = reader.get_batch(False, idx)
print filename_list
# feed data into the model
feed_dict = {
model.images: imgs
}
with tf.device(config.gpu):
#
predict_points = sess.run(predicts, feed_dict=feed_dict)
points = get_points(predict_points)
#for ii in xrange(config.points_num):
#points[0][ii * 2: ii * 2 + 2] = get_point(predict_points[0], ii)
print points
for batch_idx in xrange(config.batch_size):
with open("./labels/annos.txt", "a") as f:
f.write('0009' + str(idx * config.batch_size + batch_idx + 5781) + '.png ')
for i in xrange(config.points_num * 2):
f.write(str(int(points[batch_idx][i])) + ',')
f.write('1\n')
f.close()
def main():
config = Config()
with tf.Graph().as_default():
# create a reader object
reader = read_data.PoseReader(config.annos_path,
config.data_path, config)
# create a model object
model = cpm.CPM(config)
# feedforward
predicts = model.build_fc(False) # False: is not training
# Initializing operation
init_op = tf.global_variables_initializer()
saver = tf.train.Saver(max_to_keep=100)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
sess.run(init_op)
model.restore(sess, saver, config.load_filename)
# write the predict coordinates to file
with open(config.annos_write_path, 'w') as fw:
# start testing
for idx in xrange(config.test_num):
with tf.device("/cpu:0"):
imgs, fm, coords, begins, filename_list = \
reader.get_random_batch(distort=False)
# feed data into the model
feed_dict = {
model.images: imgs,
model.coords: coords,
model.labels: fm
}
with tf.device(config.gpu):
# run the testing operation
predict_coords_list = sess.run(predicts, feed_dict=feed_dict)
# print predict_coords_list
for filename, predict_coords in zip(filename_list, predict_coords_list):
print (filename, predict_coords)
fw.write(filename + ' ')
for i in xrange(config.points_num):
# w = predict_coords[i*2] * config.img_width
# h = predict_coords[i*2 + 1] * config.img_height
# item = str(int(w)) + ',' + str(int(h))
w = predict_coords[i*2]
h = predict_coords[i*2 + 1]
item = str(w) + ',' + str(h)
fw.write(item + ',')
fw.write('1\n')
def main():
config = Config()
with tf.Graph().as_default():
# create a reader object
reader = read_data.PoseReader("./labels/txt/validate_annos.txt",
"./data/train_imgs/", config)
# create a model object
model = cpm.CPM(config)
# feedforward
predicts = model.build_fc(False)
# return the loss
loss = model.loss()
# Initializing operation
init_op = tf.global_variables_initializer()
saver = tf.train.Saver(max_to_keep=100)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
sess.run(init_op)
model.restore(sess, saver, config.load_filename)
# start testing
for idx in xrange(config.test_num):
with tf.device("/cpu:0"):
imgs, fm, coords, begins, filename_list = reader.get_batch(
distort=False)
# feed data into the model
feed_dict = {
model.images: imgs,
model.coords: coords,
model.labels: fm
}
with tf.device(config.gpu):
#
predict_coords = sess.run(predicts, feed_dict=feed_dict)
def main():
config = Config()
with tf.Graph().as_default():
# create a reader object
reader = read_data.PoseReader(config)
# create a model object
model = cpm.CPM(config)
# feedforward
predicts = model.build_fc(True)
# return the loss
loss = model.loss()
# training operation
train_op = model.train_op(loss, model.global_step)
# Initializing operation
init_op = tf.global_variables_initializer()
saver = tf.train.Saver(max_to_keep=100)
sess_config = tf.ConfigProto(log_device_placement=True)
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
# initialize parameters or restore from previous model
if not os.path.exists(config.params_dir):
os.makedirs(config.params_dir)
if os.listdir(config.params_dir) == [] or config.initialize:
print("Initializing Network")
sess.run(init_op)
else:
sess.run(init_op)
model.restore(sess, saver, config.load_filename)
merged = tf.summary.merge_all()
logdir = os.path.join(config.logdir,
datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
writer = tf.summary.FileWriter(logdir, sess.graph)
new_start = 0
offset = 0
epoch_count = 1
print "Last ", config.image_count % config.batch_size, "will be missed in every ", config.image_count, "images"
# start training
for idx in xrange(config.max_iteration):
with tf.device(config.gpu):
if (offset + config.image_count) <= (idx *
config.batch_size):
# new_start = 0
offset = ((idx - 1) * config.batch_size)
epoch_count = epoch_count + 1
print "start of new epoch"
imgs, fm, coords, begins, filename_list,end_index = \
reader.get_random_batch( new_start, distort=config.toDistort )
new_start = end_index + 1
print filename_list
print "are the file names"
print coords
print "are the coordinates"
# feed data into the model
feed_dict = {
model.images: imgs,
model.coords: coords,
model.labels: fm
}
with tf.device(config.gpu):
# run the training operation
sess.run(train_op, feed_dict=feed_dict)
with tf.device(config.gpu):
# write summary
if (idx + 1) % config.summary_iters == 0:
print("iter#", idx + 1)
tmp_global_step = model.global_step.eval()
summary = sess.run(merged, feed_dict=feed_dict)
writer.add_summary(summary, tmp_global_step)
# save checkpoint
if (idx + 1) % config.checkpoint_iters == 0:
tmp_global_step = model.global_step.eval()
save_path = model.save(sess, saver,
config.save_filename,
tmp_global_step)
norm_image_size = (368, 368)
belief_map_size = (46, 46)
keypoints_count = len(utils.keypoints_order[category])
# 2.load data for training
train_data = utils.dataLoader(
category=category,
path_to_excel_file=
'/home/panziqi/project/fashion_ai/annotations/train/train.xlsx',
images_prefix='/home/public/FashionAI/keypoint/season1/train/',
norm_image_size=norm_image_size,
belief_map_size=belief_map_size)
# 3.train
cpm = cpm.CPM(network_name='CPM_' + category,
stage_count=4,
norm_image_size=norm_image_size,
belief_map_size=belief_map_size,
keypoints_count=keypoints_count)
cpm.train(
train_data=train_data,
log_folder='/home/panziqi/project/fashion_ai/version_softmax/log/train/',
params_folder='/home/panziqi/project/fashion_ai/version_softmax/params/',
epochs=epochs,
batch_size=batch_size,
init_lr=0.015,
lr_step=3,
lr_factor=0.7,
ctx=[mx.gpu(gpu_id) for gpu_id in range(gpu_s, gpu_e)])
print("read data from: " + file_name)
print("read checkpoint from: " + ckpt_file_name)
df = pd.read_csv(file_name)
if total_size > -1:
df = df[:total_size]
pred_df = df[["image_id", "image_category"]]
input_data = train_input.data_stream(df,
config.batch_size,
is_train=False,
scale=scale,
pre_path=pre_path)
tf.reset_default_graph()
#with tf.device('/device:GPU:0'):
model = cpm.CPM(config)
predict = model.inference_pose_vgg(is_train=False)
saver = tf.train.Saver(tf.trainable_variables())
init_op = tf.global_variables_initializer()
config_cpu = tf.ConfigProto(device_count={'GPU': 0})
with tf.Session() as sess:
sess.run(init_op)
# if not os.path.exists(config.params_dir):
# os.makedirs(config.params_dir)
# if os.listdir(config.params_dir) == [] or config.initialize:
print("Initializing Network")
def main():
config = Config()
with tf.Graph().as_default():
# create a reader object
reader = read_data.PoseReader("./labels/txt/validate_annos.txt",
"./data/train_imgs/", config)
# create a model object
model = cpm.CPM(config)
# feedforward
predicts = model.build_fc(True)
# return the loss
loss = model.loss()
# training operation
train_op = model.train_op(loss, model.global_step)
# Initializing operation
init_op = tf.global_variables_initializer()
saver = tf.train.Saver(max_to_keep=100)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
# initialize parameters or restore from previous model
if not os.path.exists(config.params_dir):
os.makedirs(config.params_dir)
if os.listdir(config.params_dir) == [] or config.initialize:
print "Initializing Network"
sess.run(init_op)
else:
sess.run(init_op)
model.restore(sess, saver, config.load_filename)
merged = tf.summary.merge_all()
logdir = os.path.join(config.logdir,
datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
writer = tf.summary.FileWriter(logdir, sess.graph)
# start training
for idx in xrange(config.max_iteration):
with tf.device("/cpu:0"):
imgs, fm, coords, begins, filename_list = reader.get_random_batch(
)
# feed data into the model
feed_dict = {
model.images: imgs,
model.coords: coords,
model.labels: fm
}
with tf.device(config.gpu):
# run the training operation
sess.run(train_op, feed_dict=feed_dict)
with tf.device('/cpu:0'):
# write summary
if (idx + 1) % config.summary_iters == 0:
tmp_global_step = model.global_step.eval()
summary = sess.run(merged, feed_dict=feed_dict)
writer.add_summary(summary, tmp_global_step)
# save checkpoint
if (idx + 1) % config.checkpoint_iters == 0:
tmp_global_step = model.global_step.eval()
model.save(sess, saver, config.save_filename,
tmp_global_step)