def __init__(self, batch_size, train_video_length_info, unit_feature_size, unit_size, lambda_reg, lr, train_clip_path, test_clip_path, train_flow_feature_dir, train_appr_feature_dir, test_flow_feature_dir, test_appr_feature_dir):
self.batch_size = batch_size
self.test_batch_size = 1
self.lr = lr
self.lambda_reg = lambda_reg
self.unit_feature_size = unit_feature_size # 4096
self.visual_feature_dim = unit_feature_size # 4096
self.train_set = TrainingDataSet(train_flow_feature_dir,train_appr_feature_dir,train_clip_path, batch_size, train_video_length_info,unit_feature_size,unit_size)
self.test_set = TestingDataSet(test_flow_feature_dir, test_appr_feature_dir, test_clip_path, self.test_batch_size)
def __init__(self, batch_size, train_csv_path, test_csv_path, test_visual_feature_dir, train_visual_feature_dir):
self.batch_size = batch_size
self.test_batch_size = 1
self.vs_lr = 0.005
self.lambda_regression = 0.01
self.alpha = 1.0/batch_size
self.semantic_size = 1024 # the size of visual and semantic comparison size
self.sentence_embedding_size = 4800
self.visual_feature_dim = 4096*3
self.train_set=TrainingDataSet(train_visual_feature_dir, train_csv_path, self.batch_size)
self.test_set=TestingDataSet(test_visual_feature_dir, test_csv_path, self.test_batch_size)
def __init__(self, config):
"""Initialization
"""
self.config = config
self.sess = None
self.saver = None
self.train_clip_path = self.config.train_clip_path
self.background_path = self.config.background_path
self.test_clip_path = self.config.test_clip_path
self.train_flow_feature_dir = self.config.train_flow_feature_dir
self.train_appr_feature_dir = self.config.train_appr_feature_dir
self.test_flow_feature_dir = self.config.test_flow_feature_dir
self.test_appr_feature_dir = self.config.test_appr_feature_dir
self.test_len_dict = self.config.test_len_dict
self.batch_size = self.config.batch_size
self.test_batch_size = 1
self.middle_layer_size = 1000
self.lambda_reg = float(self.config.lambda_reg)
self.action_class_num = self.config.action_class_num
self.feat_type = self.config.feat_type
self.visual_feature_dim = self.config.visual_feature_dim
# Initialize the training data and testing data
self.train_set = TrainingDataSet(self.config,
self.train_flow_feature_dir,
self.train_appr_feature_dir,
self.train_clip_path,
self.background_path)
self.test_set = TestingDataSet(self.config, self.test_flow_feature_dir,
self.test_appr_feature_dir,
self.test_clip_path,
self.test_batch_size,
self.test_len_dict)
# Path to save the summary of the models
self.summary_dir = os.path.join('./summary', self.config.save_name)
if not os.path.exists(self.summary_dir):
os.mkdir(self.summary_dir)
if self.config.issave == 'Yes':
self.model_dir = os.path.join('./model', self.config.save_name)
if not os.path.exists(self.model_dir):
os.mkdir(self.model_dir)
def __init__(self, batch_size, train_video_length_info, ctx_num,
unit_feature_size, unit_size, lambda_reg, lr, train_clip_path,
background_path, test_clip_path, train_visual_feature_dir,
test_visual_feature_dir):
self.batch_size = batch_size
self.test_batch_size = 1
self.lr = lr
self.lambda_reg = lambda_reg
self.unit_feature_size = unit_feature_size
self.visual_feature_dim = unit_feature_size * 3
self.train_set = TrainingDataSet(train_visual_feature_dir,
train_clip_path, background_path,
batch_size, train_video_length_info,
ctx_num, unit_feature_size, unit_size)
self.test_set = TestingDataSet(test_visual_feature_dir, test_clip_path,
self.test_batch_size, ctx_num)
def __init__(self, batch_size, pool_size, train_csv_path, test_csv_path,
test_visual_feature_dir, train_visual_feature_dir):
self.batch_size = batch_size
self.test_batch_size = 1
self.vs_lr = 0.0001
self.lambda_regression = 0.01
self.alpha = 1.0 / batch_size
#self.alpha=0.06
self.pool_size = pool_size
self.semantic_size = 1024
self.sentence_embedding_size = 4800
self.visual_feature_dim = 4096
self.train_set = TrainingDataSet(train_visual_feature_dir,
train_csv_path, self.batch_size)
self.test_set = TestingDataSet(test_visual_feature_dir, test_csv_path,
self.test_batch_size)
self.context_num = 1
def __init__(self, batch_size, ctx_num, unit_size, unit_feature_size,
action_class_num, lr, lambda_reg, train_clip_path,
background_path, test_clip_path, train_flow_feature_dir,
train_appr_feature_dir, test_flow_feature_dir,
test_appr_feature_dir):
self.batch_size = batch_size
self.test_batch_size = 1 # 测试的时候batch = 1个clip,训练的时候batcch = 128
self.middle_layer_size = 1000
self.vs_lr = lr
self.lambda_reg = lambda_reg # 1.0
self.action_class_num = action_class_num # 20
self.visual_feature_dim = unit_feature_size * 3 # 4096*3
self.train_set = TrainingDataSet(train_flow_feature_dir,
train_appr_feature_dir,
train_clip_path, background_path,
batch_size, ctx_num, unit_size,
unit_feature_size, action_class_num)
self.test_set = TestingDataSet(test_flow_feature_dir,
test_appr_feature_dir, test_clip_path,
self.test_batch_size, unit_size)
def __init__(self, batch_size, train_csv_path, test_csv_path,
test_visual_feature_dir, sliding_dir,
sliding_training_sample_file, test_clip_sentence_pairs_path,
test_swin_txt_path, train_softmax_dir, test_softmax_dir):
self.batch_size = batch_size
self.test_batch_size = 1
self.vs_lr = 0.005
self.lambda_regression = 0.01
self.alpha = 1.0 / batch_size
self.semantic_size = 1024 # the size of visual and semantic comparison size
self.action_semantic_size = 300
self.sentence_embedding_size = 4800
self.visual_feature_dim = 4096 * 3
self.train_set = TrainingDataSet(sliding_dir,
sliding_training_sample_file,
train_csv_path, batch_size,
train_softmax_dir)
self.test_set = TestingDataSet(test_visual_feature_dir, test_csv_path,
self.test_batch_size,
test_swin_txt_path, test_softmax_dir,
test_clip_sentence_pairs_path)
def __init__(self, ):
self.train_set = TrainingDataSet(self.batch_size)
self.test_set = TestingDataSet()
initial_steps = 0
max_steps = 20000
batch_size = 64
train_csv_path = "/home/wam/Action_Recognition/TACoS/train_clip-sentvec.pkl"
test_csv_path = "/home/wam/Action_Recognition/TACoS/test_clip-sentvec.pkl"
test_feature_dir="/home/wam/Action_Recognition/Interval128_256_overlap0.8_c3d_fc6/"
train_feature_dir = "/home/wam/Action_Recognition/Interval64_128_256_512_overlap0.8_c3d_fc6/"
test_batch_size = 1
vs_lr = 0.001
lambda_regression = 0.01
alpha = 1.0/batch_size
semantic_size = 1024 # the size of visual and semantic comparison size
sentence_embedding_size = 4800
visual_feature_dim = 4096*3
train_set=TrainingDataSet(train_feature_dir, train_csv_path, batch_size)
test_set=TestingDataSet(test_feature_dir, test_csv_path, test_batch_size)
def compute_loss_reg(sim_reg_mat, offset_label):
sim_score_mat, p_reg_mat, l_reg_mat = tf.split(sim_reg_mat, 3, 2)
sim_score_mat = tf.reshape(sim_score_mat, [batch_size, batch_size])
l_reg_mat = tf.reshape(l_reg_mat, [batch_size, batch_size])
p_reg_mat = tf.reshape(p_reg_mat, [batch_size, batch_size])
# unit matrix with -2
I_2 = tf.diag(tf.constant(-2.0, shape=[batch_size]))
all1 = tf.constant(1.0, shape=[batch_size, batch_size])
mask_mat = tf.add(I_2, all1)
# loss cls, not considering iou
I = tf.diag(tf.constant(1.0, shape=[batch_size]))
I_half = tf.diag(tf.constant(0.5, shape=[batch_size]))
batch_para_mat = tf.constant(alpha, shape=[batch_size, batch_size])
def __init__(self, ):
self.train_set = TrainingDataSet(self.batch_size)
self.val_set = ValidationDataSet()