def get_validation_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in [
'kinetics', 'mini-kinetics', 'activitynet', 'ucf101', 'hmdb51'
]
if opt.dataset in ['kinetics', 'mini-kinetics']:
validation_data = Kinetics(opt.video_path,
opt.annotation_path,
opt.n_frames_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
validation_data = UCF101(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
validation_data = HMDB51(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
return validation_data
def get_training_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'jester'
]
if opt.dataset == 'kinetics':
training_data = Kinetics(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'activitynet':
training_data = ActivityNet(opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'ucf101':
training_data = UCF101(opt.video_path,
opt.annotation_path,
'train',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'hmdb51':
training_data = HMDB51(opt.video_path,
opt.annotation_path,
'train',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'jester':
training_data = Jester(opt.video_path,
opt.annotation_path,
'train',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
return training_data
def get_validation_data(video_path,
annotation_path,
dataset_name,
file_type,
spatial_transform=None,
temporal_transform=None,
target_transform=None):
assert dataset_name in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'mit'
]
assert file_type in ['jpg', 'hdf5']
if file_type == 'jpg':
loader = VideoLoader(lambda x: f'image_{x:05d}.jpg')
video_path_formatter = (
lambda root_path, label, video_id: root_path / label / video_id)
else:
loader = VideoLoaderHDF5()
video_path_formatter = (lambda root_path, label, video_id: root_path /
label / f'{video_id}.hdf5')
if dataset_name == 'activitynet':
validation_data = ActivityNet(
video_path,
annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
else:
validation_data = VideoDatasetMultiClips(
video_path,
annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
return validation_data, collate_fn
def get_validation_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51']
if opt.dataset == 'kinetics':
validation_data = Kinetics(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(
opt.video_path,
opt.annotation_path,
'validation',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'ucf101':
validation_data = UCF101(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'hmdb51':
validation_data = HMDB51(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
return validation_data
def get_validation_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51', 'pandakill', 'pandakill_lstm', 'court']
if opt.dataset == 'kinetics':
validation_data = Kinetics(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(
opt.video_path,
opt.annotation_path,
'validation',
False,
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
validation_data = UCF101(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
validation_data = HMDB51(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'pandakill':
validation_data = PANDAKILL(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'court':
validation_data = COURT(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'pandakill_lstm':
validation_data = PANDAKILL(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=1)
return validation_data
def get_training_data(video_path,
annotation_path,
dataset_name,
input_type,
file_type,
spatial_transform=None,
temporal_transform=None,
target_transform=None):
assert dataset_name in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'mit'
]
assert input_type in ['rgb', 'flow']
assert file_type in ['jpg', 'hdf5']
if file_type == 'jpg':
assert input_type == 'rgb', 'flow input is supported only when input type is hdf5.'
if get_image_backend() == 'accimage':
from datasets.loader import ImageLoaderAccImage
loader = VideoLoader(image_name_formatter, ImageLoaderAccImage())
else:
loader = VideoLoader(image_name_formatter)
video_path_formatter = (
lambda root_path, label, video_id: root_path / label / video_id)
else:
if input_type == 'rgb':
loader = VideoLoaderHDF5()
else:
loader = VideoLoaderFlowHDF5()
video_path_formatter = (lambda root_path, label, video_id: root_path /
label / f'{video_id}.hdf5')
if dataset_name == 'activitynet':
training_data = ActivityNet(video_path,
annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
else:
# training_data = VideoDataset(video_path,
# annotation_path,
# 'training',
# spatial_transform=spatial_transform,
# temporal_transform=temporal_transform,
# target_transform=target_transform,
# video_loader=loader,
# video_path_formatter=video_path_formatter)
training_data = BlackAIdataset_(
video_path,
annotation_path,
'train',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
return training_data
def get_training_data(video_path,
annotation_path,
dataset_name,
input_type,
file_type,
spatial_transform=None,
temporal_transform=None,
target_transform=None,
sample_t_stride=1):
assert dataset_name in [
'kinetics', 'mini_kinetics', 'activitynet', 'ucf101', 'hmdb51', 'mit',
'breakfast', 'mini_breakfast', 'movingmnist',
'movingmnist_blackframes', 'movingmnist_longterm',
'movingmnist_motiondiff', 'movingmnist_motionsame',
'movingmnist_frequencies', 'movingmnist_frequencies_complex',
'something', 'movingmnist_static'
]
assert input_type in ['rgb', 'flow']
assert file_type in ['jpg', 'hdf5', None]
if file_type == 'jpg':
assert input_type == 'rgb', 'flow input is supported only when input type is hdf5.'
if 'movingmnist' in dataset_name:
image_name_formatter = mnist_image_name_formatter
elif 'something' in dataset_name:
image_name_formatter = something_image_name_formatter
else:
image_name_formatter = usual_image_name_formatter
if get_image_backend() == 'accimage':
from datasets.loader import ImageLoaderAccImage
loader = VideoLoader(image_name_formatter, ImageLoaderAccImage())
else:
loader = VideoLoader(image_name_formatter)
video_path_formatter = (
lambda root_path, label, video_id: root_path / label / video_id)
if 'movingmnist' in dataset_name or 'something' in dataset_name:
video_path_formatter = (
lambda root_path, label, video_id: root_path / video_id)
else:
if input_type == 'rgb':
loader = VideoLoaderHDF5()
else:
loader = VideoLoaderFlowHDF5()
if dataset_name in ['kinetics', 'mini_kinetics']:
video_path_formatter = (lambda root_path, label, video_id:
root_path / label / f'{video_id}')
else:
video_path_formatter = (lambda root_path, label, video_id:
root_path / label / f'{video_id}.hdf5')
print("video_path_formatter", video_path_formatter)
if dataset_name == 'activitynet':
training_data = ActivityNet(video_path,
annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
elif dataset_name in ['kinetics', 'mini_kinetics']:
training_data = VideoDataset(Path(
os.path.join(video_path, "h5_train_frames")),
annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
else:
print("Building VideoDataset for", dataset_name)
#print(spatial_transform)
#print(temporal_transform)
#print(loader)
training_data = VideoDataset(video_path,
annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
return training_data
def __init__(self, *args, **kwargs):
ActivityNet.__init__(self, *args, **kwargs)
DatasetTSN.__init__(self, *args, **kwargs)
def get_validation_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'kinetics_adv', 'kinetics_bkgmsk', 'kinetics_human_msk', 'kinetics_adv_msk', 'activitynet', 'ucf101', 'hmdb51', 'diving48']
if opt.dataset == 'kinetics':
validation_data = Kinetics(
opt.video_path+'/val',
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'kinetics_adv':
validation_data = Kinetics_adv(
opt.video_path+'/val',
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
place_pred_path=opt.place_pred_path,
is_place_soft_label=opt.is_place_soft)
elif opt.dataset == 'kinetics_bkgmsk':
validation_data = Kinetics_bkgmsk(
opt.video_path+'/val',
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
detection_path=opt.human_dets_path,
mask_ratio=opt.mask_ratio)
elif opt.dataset == 'kinetics_adv_msk':
validation_data_1 = Kinetics_adv(
opt.video_path+'/val',
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
place_pred_path=opt.place_pred_path,
is_place_soft_label=opt.is_place_soft)
validation_data_2 = Kinetics_human_msk(
opt.video_path+'/val',
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
detection_path=opt.human_dets_path,
mask_ratio=opt.mask_ratio)
validation_data = [validation_data_1, validation_data_2]
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(
opt.video_path,
opt.annotation_path,
'validation',
False,
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
validation_data = UCF101(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
vis=opt.vis)
elif opt.dataset == 'hmdb51':
validation_data = HMDB51(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
vis=opt.vis)
elif opt.dataset == 'diving48':
validation_data = Diving48(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
vis=opt.vis)
return validation_data
def get_validation_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51','something','fire']
if opt.dataset == 'kinetics':
validation_data = Kinetics(
opt.video_path+"/val_256",
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
opt.annotation_path = opt.annotation_path.replace("train","test")
#opt.annotation_path = opt.annotation_path + "/test_rgb_ucf101.txt"
validation_data = UCF101(
opt.video_path,
opt.annotation_path,
num_segments = opt.num_segments,
modality = opt.modality,
transform = spatial_transform,
test_mode = True)
elif opt.dataset == 'hmdb51':
opt.annotation_path = opt.annotation_path.replace("train","val")
validation_data = HMDB51(
opt.video_path,
opt.annotation_path,
num_segments = opt.num_segments,
modality = opt.modality,
transform = spatial_transform)
elif opt.dataset == 'fire':
opt.annotation_path = opt.annotation_path.replace("train","test")
validation_data = FIRE(
opt.video_path,
opt.annotation_path,
num_segments = opt.num_segments,
modality = opt.modality,
transform = spatial_transform)
elif opt.dataset == 'something':
validation_data = Something(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
return validation_data
def get_training_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51']
if opt.dataset == 'kinetics':
training_data = Kinetics(opt,
opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'activitynet':
training_data = ActivityNet(opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'ucf101':
if opt.compressed:
if opt.mult_loss:
training_data = UCF101M(opt,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
else:
training_data = UCF101R(opt,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
else:
training_data = UCF101(opt,
opt.jpeg_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'hmdb51':
if opt.compressed:
if opt.mult_loss:
training_data = HMDB51M(opt,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
else:
training_data = HMDB51R(opt,
opt.annotation_path,
'training',
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
else:
training_data = HMDB51(opt,
opt.jpeg_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
return training_data
def get_training_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'pkummd', 'pkummd_sim',
'pkummd_ed'
]
if opt.dataset == 'kinetics':
training_data = Kinetics(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'activitynet':
training_data = ActivityNet(opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'ucf101':
training_data = UCF101(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'hmdb51':
training_data = HMDB51(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'pkummd':
training_data = PKUMMD(opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == "pkummd_sim":
training_data = PKUMMD_SIM(opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == "pkummd_ed":
training_data = PKUMMD_ED(opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
return training_data
def get_validation_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'pkummd', 'pkummd_sim',
'pkummd_ed'
]
if opt.dataset == 'kinetics':
validation_data = Kinetics(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(opt.video_path,
opt.annotation_path,
'validation',
False,
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
validation_data = UCF101(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
validation_data = HMDB51(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'pkummd':
validation_data = PKUMMD(opt.video_path,
opt.annotation_path,
'validation',
False,
opt.n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'pkummd_sim':
validation_data = PKUMMD_SIM(opt.video_path,
opt.annotation_path,
'validation',
False,
opt.n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'pkummd_ed':
validation_data = PKUMMD_ED(opt.video_path,
opt.annotation_path,
'validation',
False,
opt.n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
return validation_data
def get_test_set(opt, spatial_transform, temporal_transform, target_transform):
assert opt.dataset in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'pkummd'
]
assert opt.test_subset in ['val', 'test', 'train']
if opt.test_subset == 'val':
subset = 'validation'
elif opt.test_subset == 'test':
subset = 'testing'
elif opt.test_subset == 'train':
subset = "training"
if opt.dataset == 'kinetics':
test_data = Kinetics(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
test_data = ActivityNet(opt.video_path,
opt.annotation_path,
subset,
True,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
test_data = UCF101(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
test_data = HMDB51(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'pkummd':
test_data = PKUMMD(opt.video_path,
opt.annotation_path,
subset,
True,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration,
window_size=opt.window_size,
window_stride=opt.window_stride,
scores_dump_path=(opt.root_path + "/" +
opt.scores_dump_path))
elif opt.dataset == "pkummd_sim":
raise NotImplemented()
elif opt.dataset == "pkummd_ed":
raise NotImplemented()
return test_data
def get_training_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51', 'pandakill', 'pandakill_lstm','court']
if opt.dataset == 'kinetics':
training_data = Kinetics(
opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'activitynet':
training_data = ActivityNet(
opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'ucf101':
training_data = UCF101(
opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'hmdb51':
training_data = HMDB51(
opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'pandakill':
training_data = PANDAKILL(
opt.video_path,
opt.annotation_path,
'training',
n_samples_for_each_video=16,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'court':
training_data = COURT(
opt.video_path,
opt.annotation_path,
'training',
n_samples_for_each_video=16,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'pandakill_lstm':
training_data = PANDAKILL_LSTM(
opt.video_path,
opt.annotation_path,
'training',
n_samples_for_each_video=16,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
return training_data
def get_test_set(opt, spatial_transform, temporal_transform, target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51','something','fire']
assert opt.test_subset in ['val', 'test']
if opt.test_subset == 'val':
subset = 'validation'
elif opt.test_subset == 'test':
subset = 'testing'
if opt.dataset == 'kinetics':
test_data = Kinetics(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
sample_stride=opt.sample_stride)
elif opt.dataset == 'activitynet':
test_data = ActivityNet(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
opt.annotation_path = opt.annotation_path + "/test_rgb_ucf101.txt"
test_data = UCF101(
opt.video_path,
opt.annotation_path,
num_segments = opt.num_segments,
modality = opt.modality,
transform = spatial_transform,
test_mode = True)
elif opt.dataset == 'hmdb51':
test_data = HMDB51(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'something':
test_data = Something(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'fire':
opt.annotation_path = "/DATACENTER2/wxy/workspace/senet-3d/datasets/txt/test_fire.txt"
test_data = FIRE(
opt.video_path,
opt.annotation_path,
num_segments = opt.num_segments,
modality = opt.modality,
transform = spatial_transform,
test_mode=True,
test_idx=opt.test_idx)
return test_data
def get_validation_set(opt,
spatial_transform,
temporal_transform,
target_transform=None):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51']
if opt.video_level_accuracy:
val_or_test = 'test'
if opt.dataset == 'kinetics' or opt.dataset == 'ucf101':
val_or_test = 'validation'
n_val_samples = 0
else:
val_or_test = 'validation'
n_val_samples = opt.n_val_samples
if opt.dataset == 'kinetics':
validation_data = Kinetics(opt,
opt.video_path,
opt.annotation_path,
val_or_test,
n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
frames_sequence=opt.sample_duration)
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(opt.jpeg_path,
opt.annotation_path,
val_or_test,
False,
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
if opt.compressed:
validation_data = UCF101R(opt,
opt.annotation_path,
val_or_test,
n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
else:
validation_data = UCF101(opt,
opt.jpeg_path,
opt.annotation_path,
val_or_test,
n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
if opt.compressed:
validation_data = HMDB51R(opt,
opt.annotation_path,
val_or_test,
n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
else:
validation_data = HMDB51(opt,
opt.jpeg_path,
opt.annotation_path,
val_or_test,
n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
return validation_data
def get_training_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'kinetics_adv', 'kinetics_bkgmsk', 'kinetics_adv_msk', 'activitynet', 'ucf101', 'hmdb51', 'diving48']
if opt.dataset == 'kinetics':
training_data = Kinetics(
opt.video_path+'/train',
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'kinetics_adv':
training_data = Kinetics_adv(
opt.video_path+'/train',
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
place_pred_path=opt.place_pred_path,
is_place_soft_label=opt.is_place_soft)
elif opt.dataset == 'kinetics_bkgmsk':
training_data = Kinetics_bkgmsk(
opt.video_path+'/train',
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
detection_path=opt.human_dets_path,
mask_ratio=opt.mask_ratio)
elif opt.dataset == 'kinetics_adv_msk':
training_data_1 = Kinetics_adv(
opt.video_path+'/train',
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
place_pred_path=opt.place_pred_path,
is_place_soft_label=opt.is_place_soft)
training_data_2 = Kinetics_human_msk(
opt.video_path+'/train',
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
detection_path=opt.human_dets_path,
mask_ratio=opt.mask_ratio)
training_data = [training_data_1, training_data_2]
elif opt.dataset == 'activitynet':
training_data = ActivityNet(
opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'ucf101':
training_data = UCF101(
opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'hmdb51':
training_data = HMDB51(
opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'diving48':
training_data = Diving48(
opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
return training_data
def get_validation_set(opt,
spatial_transform,
temporal_transform,
target_transform,
spatio_temporal_transform=None):
assert opt.dataset in [
'kinetics',
'activitynet',
'ucf101',
'hmdb51',
'kth',
'kth2',
'sth',
'sth_init',
'gtea',
'jester',
'ucf50',
'ucf50_color',
]
if opt.dataset == 'kinetics':
validation_data = Kinetics(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(opt.video_path,
opt.annotation_path,
'validation',
False,
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
validation_data = UCF101(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
validation_data = HMDB51(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'kth':
validation_data = KTH(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'kth2':
validation_data = KTH2(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'sth':
validation_data = Something2(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'jester':
validation_data = Jester(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'sth_init':
validation_data = Something2Init(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'gtea':
validation_data = GTEA(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
test_split=opt.test_split)
elif opt.dataset == 'ucf50':
validation_data = UCF50(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
test_split=opt.test_split)
elif opt.dataset == 'ucf50_color':
validation_data = UCF50(
opt.video_path,
opt.annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
test_split=opt.test_split)
return validation_data
def get_validation_data(video_path,
annotation_path,
dataset_name,
input_type,
file_type,
spatial_transform=None,
temporal_transform=None,
target_transform=None):
assert dataset_name in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'mit', 'somethingv2',
'somethingv1'
]
assert input_type in ['rgb', 'flow']
assert file_type in ['jpg', 'hdf5']
if 'somethingv1' in dataset_name:
formatter = sthv1_image_name_formatter
elif 'somethingv2' in dataset_name:
formatter = sthv2_image_name_formatter
else:
formatter = image_name_formatter
if file_type == 'jpg':
assert input_type == 'rgb', 'flow input is supported only when input type is hdf5.'
if get_image_backend() == 'accimage':
from datasets.loader import ImageLoaderAccImage
loader = VideoLoader(formatter, ImageLoaderAccImage())
else:
loader = VideoLoader(formatter)
video_path_formatter = (
lambda root_path, label, video_id: root_path / label / video_id)
else:
if input_type == 'rgb':
loader = VideoLoaderHDF5()
else:
loader = VideoLoaderFlowHDF5()
video_path_formatter = (lambda root_path, label, video_id: root_path /
label / '{}.hdf5'.format(video_id))
if dataset_name == 'activitynet':
validation_data = ActivityNet(
video_path,
annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
else:
validation_data = VideoDatasetMultiClips(
video_path,
annotation_path,
'validation',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter)
return validation_data, collate_fn
def get_test_set(opt,
spatial_transform,
temporal_transform,
target_transform,
spatio_temporal_transform=None):
assert opt.dataset in [
'kinetics',
'activitynet',
'ucf101',
'hmdb51',
'kth',
'kth2',
'sth',
'sth_init',
'gtea',
'jester',
'ucf50',
'ucf50_color',
'real',
]
assert opt.test_subset in ['val', 'test']
if opt.test_subset == 'val':
subset = 'validation'
elif opt.test_subset == 'test':
subset = 'testing'
if opt.dataset == 'kinetics':
test_data = Kinetics(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
test_data = ActivityNet(opt.video_path,
opt.annotation_path,
subset,
True,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
test_data = UCF101(opt.video_path,
opt.annotation_path,
subset,
10,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
test_data = HMDB51(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'kth':
test_data = KTH(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'kth2':
test_data = KTH2(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'sth':
test_data = Something2(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'jester':
test_data = Jester(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'sth_init':
test_data = Something2Init(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'gtea':
test_data = GTEA(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration,
test_split=opt.test_split)
elif opt.dataset == 'ucf50':
test_data = UCF50(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration,
test_split=opt.test_split)
elif opt.dataset == 'ucf50_color':
test_data = UCF50(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration,
test_split=opt.test_split)
elif opt.dataset == 'real':
test_data = REAL(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
spatio_temporal_transform=spatio_temporal_transform,
target_transform=target_transform,
sample_duration=opt.sample_duration)
return test_data
def get_inference_data(video_path,
annotation_path,
dataset_name,
input_type,
file_type,
inference_subset,
spatial_transform=None,
temporal_transform=None,
target_transform=None,
sample_t_stride=1):
assert dataset_name in [
'kinetics', 'mini_kinetics', 'activitynet', 'ucf101', 'hmdb51', 'mit',
'breakfast', 'mini_breakfast', 'movingmnist',
'movingmnist_blackframes', 'movingmnist_longterm',
'movingmnist_motiondiff', 'movingmnist_motionsame',
'movingmnist_frequencies', 'movingmnist_frequencies_complex',
'something', 'movingmnist_static'
]
assert input_type in ['rgb', 'flow']
assert file_type in ['jpg', 'hdf5', None]
assert inference_subset in ['train', 'val', 'test']
if file_type == 'jpg':
assert input_type == 'rgb', 'flow input is supported only when input type is hdf5.'
if 'movingmnist' in dataset_name:
image_name_formatter = mnist_image_name_formatter
elif 'something' in dataset_name:
image_name_formatter = something_image_name_formatter
else:
image_name_formatter = usual_image_name_formatter
if get_image_backend() == 'accimage':
from datasets.loader import ImageLoaderAccImage
loader = VideoLoader(image_name_formatter, ImageLoaderAccImage())
else:
loader = VideoLoader(image_name_formatter)
video_path_formatter = (
lambda root_path, label, video_id: root_path / label / video_id)
if dataset_name in [
'movingmnist', 'movingmnist_blackframes',
'movingmnist_longterm', 'something'
]:
video_path_formatter = (
lambda root_path, label, video_id: root_path / video_id)
else:
if input_type == 'rgb':
loader = VideoLoaderHDF5()
else:
loader = VideoLoaderFlowHDF5()
video_path_formatter = (lambda root_path, label, video_id: root_path /
label / f'{video_id}.hdf5')
if inference_subset == 'train':
subset = 'training'
elif inference_subset == 'val':
subset = 'validation'
elif inference_subset == 'test':
subset = 'testing'
if dataset_name == 'activitynet':
inference_data = ActivityNet(video_path,
annotation_path,
subset,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter,
is_untrimmed_setting=True)
else:
inference_data = VideoDatasetMultiClips(
video_path,
annotation_path,
subset,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter,
target_type=['label', 'video_id', 'segment'])
return inference_data, collate_fn
def get_test_set(opt,
spatial_transform,
temporal_transform,
target_transform,
score_sens_mode=False,
score_inf_mode=False,
inner_temp_transform=None):
assert opt.dataset in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'something'
]
assert opt.test_subset in ['val', 'test']
if opt.test_subset == 'val':
subset = 'validation'
elif opt.test_subset == 'test':
subset = 'testing'
if opt.dataset == 'kinetics':
test_data = Kinetics(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
sample_stride=opt.sample_stride)
elif opt.dataset == 'activitynet':
test_data = ActivityNet(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
test_data = UCF101(
opt.video_path,
opt.annotation_path,
subset,
# 0,
1, # sample 1 clip each video
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
score_sens_mode=score_sens_mode,
score_inf_mode=score_inf_mode)
elif opt.dataset == 'hmdb51':
test_data = HMDB51(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'something':
test_data = Something(
opt.video_path,
opt.annotation_path,
subset,
# 0,
1, # sample 1 clip each video
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
score_sens_mode=score_sens_mode,
score_inf_mode=score_inf_mode,
inner_temp_transform=inner_temp_transform)
return test_data
def get_inference_data(video_path,
annotation_path,
dataset_name,
input_type,
file_type,
inference_subset,
spatial_transform=None,
temporal_transform=None,
target_transform=None):
assert dataset_name in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'mit'
]
assert input_type in ['rgb', 'flow']
assert file_type in ['jpg', 'hdf5']
assert inference_subset in ['train', 'val', 'test']
if file_type == 'jpg':
assert input_type == 'rgb', 'flow input is supported only when input type is hdf5.'
if get_image_backend() == 'accimage':
from datasets.loader import ImageLoaderAccImage
loader = VideoLoader(image_name_formatter, ImageLoaderAccImage())
else:
loader = VideoLoader(image_name_formatter)
video_path_formatter = (
lambda root_path, label, video_id: root_path / label / video_id)
else:
if input_type == 'rgb':
loader = VideoLoaderHDF5()
else:
loader = VideoLoaderFlowHDF5()
video_path_formatter = (lambda root_path, label, video_id: root_path /
label / f'{video_id}.hdf5')
if inference_subset == 'train':
subset = 'training'
elif inference_subset == 'val':
subset = 'validation'
elif inference_subset == 'test':
subset = 'testing'
if dataset_name == 'activitynet':
inference_data = ActivityNet(video_path,
annotation_path,
subset,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter,
is_untrimmed_setting=True)
else:
inference_data = VideoDatasetMultiClips(
video_path,
annotation_path,
subset,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform,
video_loader=loader,
video_path_formatter=video_path_formatter,
target_type=['video_id', 'segment'])
return inference_data, collate_fn
def get_test_set(opt, spatial_transform, temporal_transform, target_transform,
image_type):
assert opt.dataset in [
'kinetics', 'activitynet', 'ucf101', 'hmdb51', 'synthetic', 'vlog'
]
assert opt.test_subset in ['val', 'test']
if opt.test_subset == 'val':
subset = 'validation'
elif opt.test_subset == 'test':
subset = 'testing'
if opt.dataset == 'kinetics':
test_data = Kinetics(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
test_data = ActivityNet(opt.video_path,
opt.annotation_path,
subset,
True,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
test_data = UCF101(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
test_data = HMDB51(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'synthetic':
test_data = Synthetic(opt.video_path,
opt.annotation_path,
subset,
1,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
channel_fuse_step=opt.channel_fuse,
image_type=image_type)
elif opt.dataset == 'vlog':
test_data = Vlog(opt.video_path,
opt.annotation_path,
subset,
1,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
image_type=image_type)
return test_data
def get_test_set(opt, spatial_transform, temporal_transform, target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51', 'grit']
assert opt.test_subset in ['val', 'test']
if opt.test_subset == 'val':
subset = 'validation'
elif opt.test_subset == 'test':
subset = 'testing'
if opt.dataset == 'kinetics':
test_data = Kinetics(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
test_data = ActivityNet(
opt.video_path,
opt.annotation_path,
subset,
True,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
test_data = UCF101(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
test_data = HMDB51(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'grit':
test_data = GRIT(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
return test_data
def get_test_set(opt, spatial_transform, temporal_transform, target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51', 'pandakill', 'court']
assert opt.test_subset in ['val', 'test']
if opt.test_subset == 'val':
subset = 'validation'
elif opt.test_subset == 'test':
subset = 'validation'
if opt.dataset == 'kinetics':
test_data = Kinetics(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
test_data = ActivityNet(
opt.video_path,
opt.annotation_path,
subset,
True,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
test_data = UCF101(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
test_data = HMDB51(
opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'pandakill':
test_data = PANDAKILL(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'court':
test_data = COURT(
opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=1)
return test_data
