def get_modules(samples, path_py, path_cy, ignore_error=False):
samples = samples.split(",")
samples = list(map(int, samples))
samples.sort()
if path_py != "":
module_name_py = path_py.split("/")[-1].replace(".py", "")
module_path_py = "//".join(path_py.split("/")[:-1])
spec_py = spec_from_file_location(module_name_py, path_py)
module_py = module_from_spec(spec_py)
sys.path.insert(0, module_path_py)
spec_py.loader.exec_module(module_py)
timer_py = Timer(module_py, spec_py)
timer_py(samples, ignore_error)
if path_cy != "":
module_name_cy = path_cy.split("/")[-1].replace(
".cp36-win_amd64.pyd", "")
module_path_cy = "//".join(path_cy.split("/")[:-1])
spec_cy = spec_from_file_location(module_name_cy, path_cy)
module_cy = module_from_spec(spec_cy)
sys.path.insert(0, module_path_cy)
spec_cy.loader.exec_module(module_cy)
timer_cy = Timer(module_cy, spec_cy)
timer_cy(samples, ignore_error)
def post(self):
checked_request = self.__check_request(request.form)
if not Spot.find_by(forsquare_id=checked_request['foursquare_id']):
checked_request = self.__re_check_request(request.form)
spot = Spot(forsquare_id=checked_request['foursquare_id'],
name=checked_request['_location'],
latlng='POINT(' + checked_request['lng'] + ' ' +
checked_request['lat'] + ")")
spot.insert()
spot = Spot.find_by(forsquare_id=checked_request['foursquare_id'])
with Timer.transaction():
timer = Timer(user_id=g.user.id,
spot_id=spot.id,
start_at=checked_request["start_at"])
timer.insert()
return jsonify(status=200,
message='ok',
request=request.form,
response={'timer_id': timer.id})
def _timer_initialize(self, complexity):
self._timer = Timer(complexity['timer'], self._nightmare_mode,
self._view.timer_update, self._game_over)
self._timer.start()
def train():
total_timer = Timer()
train_timer = Timer()
load_timer = Timer()
max_epoch = 30
epoch_step = int(cfg.train_num//cfg.batch_size)
t = 1
for epoch in range(1, max_epoch + 1):
print('-'*25, 'epoch', epoch,'/',str(max_epoch), '-'*25)
t_loss = 0
ll_loss = 0
r_loss = 0
c_loss = 0
for step in range(1, epoch_step + 1):
t = t + 1
total_timer.tic()
load_timer.tic()
images, labels, imnm, num_boxes, imsize = data.get()
# load_timer.toc()
feed_dict = {input_: images,
get_boxes: labels[..., 1:5][:, ::-1, :],
get_classes: labels[..., 0].reshape((cfg.batch_size, -1))[:, ::-1]
}
if cfg.cnt_branch:
_, g_step_, tt_loss, cl_loss, cn_loss, re_loss, lr_ = sess.run(
[train_op,
global_step,
total_loss,
cls_loss,
cnt_loss,
reg_loss,
lr], feed_dict = feed_dict)
else:
_, g_step_, tt_loss, cl_loss, re_loss, lr_ = sess.run(
[train_op,
global_step,
total_loss,
cls_loss,
reg_loss,
lr], feed_dict = feed_dict)
total_timer.toc()
if g_step_%50 ==0:
sys.stdout.write('\r>> ' + 'iters '+str(g_step_)+str('/')+str(epoch_step*max_epoch)+' loss '+str(tt_loss) + ' ')
sys.stdout.flush()
summary_str = sess.run(summary_op, feed_dict = feed_dict)
train_total_summary = tf.Summary(value=[
tf.Summary.Value(tag="config/learning rate", simple_value=lr_),
tf.Summary.Value(tag="train/classification/focal_loss", simple_value=cfg.class_weight*cl_loss),
tf.Summary.Value(tag="train/classification/cnt_loss", simple_value=cfg.cnt_weight*cn_loss),
# tf.Summary.Value(tag="train/p_nm", simple_value=p_nm_),
tf.Summary.Value(tag="train/regress_loss", simple_value=cfg.regress_weight*re_loss),
# tf.Summary.Value(tag="train/clone_loss", simple_value=cfg.class_weight*cl_loss + cfg.regress_weight*re_loss + cfg.cnt_weight*cn_loss),
# tf.Summary.Value(tag="train/l2_loss", simple_value=l2_loss),
tf.Summary.Value(tag="train/total_loss", simple_value=tt_loss)
])
print('curent speed: ', total_timer.diff, 'remain time: ', total_timer.remain(g_step_, epoch_step*max_epoch))
summary_writer.add_summary(summary_str, g_step_)
summary_writer.add_summary(train_total_summary, g_step_)
if g_step_%10000 == 0:
print('saving checkpoint')
saver.save(sess, cfg.ckecpoint_file + '/model.ckpt', g_step_)
total_timer.toc()
sys.stdout.write('\n')
print('>> mean loss', t_loss)
print('curent speed: ', total_timer.average_time, 'remain time: ', total_timer.remain(g_step_, epoch_step*max_epoch))
print('saving checkpoint')
saver.save(sess, cfg.ckecpoint_file + '/model.ckpt', g_step_)
print('epoch', epoch)
g_list = tf.global_variables()
sess = tf.Session()
summary_writer = tf.summary.FileWriter(cfg.ckecpoint_file, sess.graph)
sess.run(tf.global_variables_initializer())
if restore_path is not None:
print('Restoring weights from: ' + restore_path)
restorer = tf.train.Saver(g_list)
restorer.restore(sess, restore_path)
if __name__ == '__main__':
val_timer = Timer()
data = pascal_voc('test', False, cfg.test_img_path, cfg.test_label_path,
cfg.test_img_txt, False)
val_pred = []
# saved_pred = {}
gt_dict = {}
val_rloss = 0
val_closs = 0
val_cnt_loss = 0
for val_step in range(1, cfg.test_num + 1):
val_timer.tic()
images, labels, imnm, num_boxes, imsize = data.get()
feed_dict = {
input_: images,
get_boxes: labels[..., 1:5][:, ::-1, :],
get_classes: labels[..., 0].reshape((cfg.batch_size, -1))[:, ::-1]
