################################################################################################################################################
args, unknown = parser.parse_known_args()
actions_dict = read_mapping_dict(args.mapping_file)
nClasses = len(actions_dict)
file_ptr = open(args.vid_list_file, 'r')
list_of_videos = file_ptr.read().split('\n')[1:-1]
################
# Training #####
################
if args.action == "train":
model = None
batch_gen = Base_batch_generator()
beta_frame = int(math.ceil(float(args.beta) * 30))
S_enc_frame = (int(args.S_enc) * beta_frame)
S_ant_frame = (int(args.S_ant) * beta_frame)
if args.model == "rnn":
model = ModelRNN(nClasses, args.rnn_size, args.max_seq_sz,
args.num_layers)
batch_gen = RNN_batch_generator(nClasses, args.n_iterations,
args.max_seq_sz, actions_dict,
args.alpha, S_enc_frame, S_ant_frame,
beta_frame)
elif args.model == "cnn":
model = ModelCNN(args.nRows, nClasses)
batch_gen = CNN_batch_generator(args.nRows, nClasses, actions_dict)
file_ptr = open(args.vid_list_file, 'r')
list_of_videos = file_ptr.read().split('\n')[1:-1]
list_of_fisher_vectors = None
if args.fisher_list_file is not None:
with open(args.fisher_list_file, mode='r') as f:
list_of_fisher_vectors = f.read().split('\n')[1:-1]
list_of_videos, list_of_fisher_vectors = filter_lists(
list_of_videos, list_of_fisher_vectors)
################
# Training #####
################
if args.action == "train":
model = None
batch_gen = Base_batch_generator()
if args.model == "rnn":
model = ModelRNN(nClasses, args.rnn_size, args.max_seq_sz,
args.num_layers)
batch_gen = RNN_batch_generator(nClasses, args.n_iterations,
args.max_seq_sz, actions_dict,
args.alpha)
elif args.model == "cnn":
if args.fisher_list_file is not None:
model = ModelCNN(nRows=args.nRows, nCols=64, nClasses=nClasses)
batch_gen = CNNFisherBatchGen(num_rows=args.nRows,
num_classes=nClasses,
action_to_id=actions_dict)
else:
model = ModelCNN(nRows=args.nRows, nCols=nClasses)