def __init__(self,
dataset_dict,
single_label=False,
random_select=False,
max_len=None):
self.dataset_dict = dataset_dict
self.single_label = single_label
self.random_select = random_select
self.max_len = max_len
if self.single_label:
self.dataset_dict = get_single_label_dict(self.dataset_dict)
self.video_list = list(self.dataset_dict.keys())
self.video_list.sort()
def __init__(self,
dataset_dict,
single_label=False,
random_select=False,
max_len=None):
self.dataset_dict = dataset_dict
self.single_label = single_label
self.random_select = random_select
self.max_len = max_len
# some videos may have more than one labels, if single_label==True, will make
# these videos as multi videos with single labels, but the weight will be reduced
if self.single_label:
self.dataset_dict = get_single_label_dict(self.dataset_dict)
self.video_list = list(self.dataset_dict.keys())
self.video_list.sort()
sample_rate=sample_rate,
base_sample_rate=base_sample_rate,
action_class_num=action_class_num,
modality='both',
feature_type=feature_type,
feature_oversample=feature_oversample,
temporal_aug=True,
load_background=False)
test_test_dataset = SingleVideoDataset(test_dataset_dict,
single_label=True,
random_select=False,
max_len=None)
test_test_loader = torch.utils.data.DataLoader(test_test_dataset,
batch_size=1,
pin_memory=True,
shuffle=False)
else:
test_test_loader = None
single_label_train_dict = get_single_label_dict(train_dataset_dict)
label_2_video = get_label_2_video(single_label_train_dict)
num_of_video = len(single_label_train_dict)
for run_idx in range(train_run_num):
naming = '{}-run-{}'.format(experiment_naming, run_idx)
my_train(train_train_dataset, train_test_loader, test_test_loader,
'rgb', naming, label_2_video, num_of_video)