Y_action = tf.placeholder(tf.float32, shape=(batch_size, tscale))
Y_start = tf.placeholder(tf.float32, shape=(batch_size, tscale))
Y_end = tf.placeholder(tf.float32, shape=(batch_size, tscale))
Y_iou = tf.placeholder(tf.float32, shape=(batch_size, tscale, tscale, 1))
LR = tf.placeholder(tf.float32)
train = tf.placeholder(tf.bool)
optimizer, loss, DBG_trainable_variables = \
DBG_Train(X_feature, Y_action, Y_start, Y_end, Y_iou, LR, train)
""" init tf"""
model_saver = tf.train.Saver(max_to_keep=80)
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
sess = tf.InteractiveSession(config=tf_config)
tf.global_variables_initializer().run()
""" load dataset """
train_dict, val_dict, test_dict = data_loader.getDatasetDict(
dbg_config.video_info_file, dbg_config.video_filter)
train_data_dict, train_video_mean_len = data_loader.getFullData(
train_dict, dbg_config)
val_data_dict = data_loader.getFullData(val_dict,
dbg_config,
training=False)
train_info = {
"cost": [],
"loss_action": [],
"loss_start": [],
"loss_end": [],
"l2": [],
'loss_iou': [],
'loss_st': []
}
prop_start = prop_start * tf_mask
prop_end = prop_end * tf_mask
# boundary map fusion
pstart = tf.reduce_sum(prop_start, 2) / tf.maximum(tf.reduce_sum(tf_mask, 2), 1)
pend = tf.reduce_sum(prop_end, 1) / tf.maximum(tf.reduce_sum(tf_mask, 1), 1)
model_saver = tf.train.Saver()
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
sess = tf.InteractiveSession(config=tf_config)
tf.global_variables_initializer().run()
model_saver.restore(sess, os.path.join(checkpoint_dir, 'dbg_model_best'))
train_dict, val_dict, test_dict = data_loader.getDatasetDict()
if test_mode == 'validation':
video_dict = val_dict
else:
video_dict = test_dict
batch_video_list = data_loader.getBatchListTest(video_dict, batch_size, shuffle=False)
batch_result_xmin = []
batch_result_xmax = []
batch_result_iou = []
batch_result_pstart = []
batch_result_pend = []
print('Runing DBG model ...')
for idx in tqdm.tqdm(range(len(batch_video_list))):
prop_start = prop_start * tf_mask
prop_end = prop_end * tf_mask
pstart = tf.reduce_sum(prop_start, 2) / tf.maximum(
tf.reduce_sum(tf_mask, 2), 1)
pend = tf.reduce_sum(prop_end, 1) / tf.maximum(tf.reduce_sum(tf_mask, 1),
1)
""" Load model weights """
model_saver = tf.train.Saver()
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
sess = tf.InteractiveSession(config=tf_config)
tf.global_variables_initializer().run()
model_saver.restore(sess, os.path.join(checkpoint_dir, 'dbg_model_best'))
""" Get test or validation video list
"""
train_dict, val_dict, test_dict = data_loader.getDatasetDict(
video_info_file)
if test_mode == 'validation':
video_dict = val_dict
else:
video_dict = test_dict
""" load test or validation data
"""
batch_video_list = data_loader.getBatchListTest(video_dict, batch_size)
""" init result list
"""
batch_result_xmin = []
batch_result_xmax = []
batch_result_iou = []
batch_result_pstart = []
batch_result_pend = []
""" Run DBG model