def get_inference_utils(opt):
assert opt.inference_crop in ['center', 'nocrop']
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [Resize(opt.sample_size)]
if opt.inference_crop == 'center':
spatial_transform.append(CenterCrop(opt.sample_size))
spatial_transform.append(ToArray())
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
SlidingWindow(opt.sample_duration, opt.inference_stride))
temporal_transform = TemporalCompose(temporal_transform)
inference_data = get_inference_data(opt.video_path, opt.annotation_path,
opt.dataset, opt.input_type,
opt.file_type, opt.inference_subset,
spatial_transform, temporal_transform)
inference_loader = paddle.batch(inference_data.reader,
batch_size=opt.inference_batch_size)
return inference_loader, inference_data.class_names
def get_inference_utils(opt):
assert opt.inference_crop in ['center', 'nocrop']
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [Resize(opt.sample_size)]
if opt.inference_crop == 'center':
spatial_transform.append(CenterCrop(opt.sample_size))
spatial_transform.extend(
[ToTensor(), ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
SlidingWindow(opt.sample_duration, opt.inference_stride))
temporal_transform = TemporalCompose(temporal_transform)
inference_data, collate_fn = get_inference_data(
opt.video_path, opt.annotation_path, opt.dataset, opt.file_type,
opt.inference_subset, spatial_transform, temporal_transform)
inference_loader = torch.utils.data.DataLoader(
inference_data,
batch_size=opt.inference_batch_size,
shuffle=False,
num_workers=opt.n_threads,
pin_memory=True,
worker_init_fn=worker_init_fn,
collate_fn=collate_fn)
return inference_loader, inference_data.class_names
def get_inference_utils(opt):
assert opt.inference_crop in ['center', 'nocrop']
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [Resize(opt.sample_size)]
if opt.inference_crop == 'center':
spatial_transform.append(CenterCrop(opt.sample_size))
spatial_transform.append(ToTensor())
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
SlidingWindow(opt.sample_duration, opt.inference_stride))
temporal_transform = TemporalCompose(temporal_transform)
inference_data, collate_fn = get_inference_data(opt.inference_label_path, opt.video_id_path,
'test', opt.inference_frame_dir, opt.image_size, window_size=opt.window_size)
inference_loader = torch.utils.data.DataLoader(
inference_data,
batch_size=opt.inference_batch_size,
shuffle=False,
num_workers=opt.n_threads,
pin_memory=False,
worker_init_fn=worker_init_fn)
# collate_fn=collate_fn)
return inference_loader, inference_data.class_names
def get_val_utils(opt):
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [
Resize(opt.sample_size),
CenterCrop(opt.sample_size),
ToArray(),
]
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
TemporalEvenCrop(opt.sample_duration, opt.n_val_samples))
temporal_transform = TemporalCompose(temporal_transform)
val_data = get_validation_data(opt.video_path, opt.annotation_path,
opt.dataset, opt.input_type, opt.file_type,
spatial_transform, temporal_transform)
val_loader = paddle.batch(val_data.reader, batch_size=opt.batch_size)
val_logger = Logger(opt.result_path / 'val.log', ['epoch', 'loss', 'acc'])
return val_loader, val_logger
def get_val_utils(opt):
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
if opt.train_crop == 'other':
spatial_transform = [
Resize((opt.scale_h, opt.scale_w)),
RandomCrop(opt.sample_size),
ToTensor()
]
else:
spatial_transform = [
Resize(opt.sample_size),
CenterCrop(opt.sample_size),
ToTensor()
]
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
TemporalEvenCrop(opt.sample_duration, opt.n_val_samples))
temporal_transform = TemporalCompose(temporal_transform)
val_data, collate_fn = get_validation_data(
opt.video_path, opt.annotation_path, opt.dataset, opt.input_type,
opt.file_type, spatial_transform, temporal_transform)
if opt.distributed:
val_sampler = torch.utils.data.distributed.DistributedSampler(
val_data, shuffle=False)
else:
val_sampler = None
val_loader = torch.utils.data.DataLoader(val_data,
batch_size=(opt.batch_size //
opt.n_val_samples),
shuffle=False,
num_workers=opt.n_threads,
pin_memory=True,
sampler=val_sampler,
worker_init_fn=worker_init_fn,
collate_fn=collate_fn)
if opt.is_master_node:
val_logger = Logger(opt.result_path / 'val.log',
['epoch', 'loss', 'acc', 'acc_num'])
else:
val_logger = None
return val_loader, val_logger
def get_inference_utils(opt):
assert opt.inference_crop in ['center', 'nocrop']
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [Resize(opt.sample_size)]
if opt.inference_crop == 'center':
spatial_transform.append(CenterCrop(opt.sample_size))
spatial_transform.append(ToTensor())
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
SlidingWindow(opt.sample_duration, opt.inference_stride))
temporal_transform = TemporalCompose(temporal_transform)
inf_data_checkpoint_path = opt.result_path / Path('inf_data_' +
opt.dataset + '.data')
inf_collate_checkpoint_path = opt.result_path / Path('inf_coll_' +
opt.dataset + '.data')
if os.path.exists(inf_data_checkpoint_path) and os.path.exists(
inf_collate_checkpoint_path) and opt.save_load_data_checkpoint:
with open(inf_data_checkpoint_path, 'rb') as filehandle:
inference_data = pickle.load(filehandle)
with open(inf_collate_checkpoint_path, 'rb') as filehandle:
collate_fn = pickle.load(filehandle)
else:
inference_data, collate_fn = get_inference_data(
opt.video_path, opt.annotation_path, opt.dataset, opt.input_type,
opt.file_type, opt.inference_subset, spatial_transform,
temporal_transform)
if opt.save_load_data_checkpoint:
with open(inf_data_checkpoint_path, 'wb') as filehandle:
pickle.dump(inference_data, filehandle)
with open(inf_collate_checkpoint_path, 'wb') as filehandle:
pickle.dump(collate_fn, filehandle)
inference_loader = torch.utils.data.DataLoader(
inference_data,
batch_size=opt.inference_batch_size,
shuffle=False,
num_workers=opt.n_threads,
pin_memory=True,
worker_init_fn=worker_init_fn,
collate_fn=collate_fn)
return inference_loader, inference_data.class_names
def compute_saliency_maps(model, opt):
# Generate tiny data loader
# Loop through it to generate saliency maps
assert opt.inference_crop in ['center', 'nocrop']
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [Resize(opt.sample_size)]
if opt.inference_crop == 'center':
spatial_transform.append(CenterCrop(opt.sample_size))
spatial_transform.append(ToTensor())
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
SlidingWindow(opt.sample_duration, opt.inference_stride))
temporal_transform = TemporalCompose(temporal_transform)
tiny_video_path = Path('/home/ruta/teeny_data/nturgb/jpg')
tiny_annotation_path = Path('/home/ruta/teeny_data/ntu_01.json')
tiny_data, collate_fn = get_inference_data(
tiny_video_path, tiny_annotation_path, opt.dataset, opt.input_type,
opt.file_type, opt.inference_subset, spatial_transform,
temporal_transform)
tiny_loader = torch.utils.data.DataLoader(
tiny_data,
batch_size=opt.inference_batch_size,
shuffle=False,
num_workers=opt.n_threads,
pin_memory=True,
sampler=None,
worker_init_fn=worker_init_fn,
collate_fn=collate_fn)
saliency_maps = []
for i, (inputs, targets) in enumerate(tiny_loader):
sal_map = get_saliency_map(inputs, targets, model, opt)
# Plot the saliency map using matplotlib and save to a file
plot_saliency(sal_map, i, inputs, targets)
saliency_maps.append(sal_map)
return saliency_maps
def get_inference_utils(opt):
assert opt.inference_crop in ['center', 'nocrop']
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [Resize(opt.sample_size)]
if opt.inference_crop == 'center':
spatial_transform.append(CenterCrop(opt.sample_size))
spatial_transform.append(ToTensor())
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
SlidingWindow(opt.sample_duration, opt.inference_stride))
temporal_transform = TemporalCompose(temporal_transform)
inference_data, collate_fn = get_inference_data(opt.video_path,
opt.input_type,
opt.file_type,
spatial_transform,
temporal_transform)
# inference_data, collate_fn = get_inference_data(
# opt.video_path, opt.input_type, opt.file_type,
# spatial_transform)
inference_loader = torch.utils.data.DataLoader(
inference_data,
batch_size=opt.inference_batch_size,
shuffle=False,
num_workers=opt.n_threads,
pin_memory=True,
worker_init_fn=worker_init_fn,
collate_fn=collate_fn)
df = pd.read_csv('kinetics_700_labels.csv')
class_names = {}
for i in range(df.shape[0]):
row = df.iloc[i]
class_names[row[0]] = row[1]
return inference_loader, class_names
def retrieve_spatial_temporal_transforms(opt):
opt.mean, opt.std = get_mean_std(opt.value_scale, dataset=opt.mean_dataset)
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [Resize(opt.sample_size),
CenterCrop(opt.sample_size),
ToTensor(),
ScaleValue(opt.value_scale),
normalize]
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
TemporalNonOverlappingWindow(opt.sample_duration))
temporal_transform = TemporalCompose(temporal_transform)
return spatial_transform, temporal_transform
def get_val_utils(opt):
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [
Resize(opt.sample_size),
CenterCrop(opt.sample_size),
ToTensor()
]
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
TemporalEvenCrop(opt.sample_duration, opt.n_val_samples))
temporal_transform = TemporalCompose(temporal_transform)
val_data, collate_fn = get_validation_data(opt.label_path, opt.video_id_path,
'val', opt.frame_dir, opt.image_size, window_size=opt.window_size)
if opt.distributed:
val_sampler = torch.utils.data.distributed.DistributedSampler(
val_data, shuffle=False)
else:
val_sampler = None
val_loader = torch.utils.data.DataLoader(val_data,
batch_size=(opt.batch_size //
opt.n_val_samples),
shuffle=False,
num_workers=opt.n_threads,
pin_memory=False,
sampler=val_sampler,
worker_init_fn=worker_init_fn)
# collate_fn=collate_fn)
if opt.is_master_node:
val_logger = Logger(opt.result_path / 'val.log',
['epoch', 'loss', 'acc', 'precision', 'recall', 'f1', 'tiou'])
else:
val_logger = None
return val_loader, val_logger
def get_train_utils(opt, model_parameters):
assert opt.train_crop in ['random', 'corner', 'center']
spatial_transform = []
if opt.train_crop == 'random':
spatial_transform.append(
RandomResizedCrop(
opt.sample_size, (opt.train_crop_min_scale, 1.0),
(opt.train_crop_min_ratio, 1.0 / opt.train_crop_min_ratio)))
elif opt.train_crop == 'corner':
scales = [1.0]
scale_step = 1 / (2**(1 / 4))
for _ in range(1, 5):
scales.append(scales[-1] * scale_step)
spatial_transform.append(MultiScaleCornerCrop(opt.sample_size, scales))
elif opt.train_crop == 'center':
spatial_transform.append(Resize(opt.sample_size))
spatial_transform.append(CenterCrop(opt.sample_size))
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
if not opt.no_hflip:
spatial_transform.append(RandomHorizontalFlip())
if opt.colorjitter:
spatial_transform.append(ColorJitter())
spatial_transform.append(ToTensor())
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.append(ScaleValue(opt.value_scale))
spatial_transform.append(normalize)
spatial_transform = Compose(spatial_transform)
assert opt.train_t_crop in ['random', 'center']
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
if opt.train_t_crop == 'random':
temporal_transform.append(TemporalRandomCrop(opt.sample_duration))
elif opt.train_t_crop == 'center':
temporal_transform.append(TemporalCenterCrop(opt.sample_duration))
temporal_transform = TemporalCompose(temporal_transform)
train_data = get_training_data(opt.video_path, opt.annotation_path,
opt.dataset, opt.input_type, opt.file_type,
spatial_transform, temporal_transform)
if opt.distributed:
train_sampler = torch.utils.data.distributed.DistributedSampler(
train_data)
else:
train_sampler = None
train_loader = torch.utils.data.DataLoader(train_data,
batch_size=opt.batch_size,
shuffle=(train_sampler is None),
num_workers=opt.n_threads,
pin_memory=True,
sampler=train_sampler,
worker_init_fn=worker_init_fn)
if opt.is_master_node:
train_logger = Logger(opt.result_path / 'train.log',
['epoch', 'loss', 'acc', 'lr'])
train_batch_logger = Logger(
opt.result_path / 'train_batch.log',
['epoch', 'batch', 'iter', 'loss', 'acc', 'lr'])
else:
train_logger = None
train_batch_logger = None
if opt.nesterov:
dampening = 0
else:
dampening = opt.dampening
optimizer = SGD(model_parameters,
lr=opt.learning_rate,
momentum=opt.momentum,
dampening=dampening,
weight_decay=opt.weight_decay,
nesterov=opt.nesterov)
assert opt.lr_scheduler in ['plateau', 'multistep']
assert not (opt.lr_scheduler == 'plateau' and opt.no_val)
if opt.lr_scheduler == 'plateau':
scheduler = lr_scheduler.ReduceLROnPlateau(
optimizer, 'min', patience=opt.plateau_patience)
else:
scheduler = lr_scheduler.MultiStepLR(optimizer,
opt.multistep_milestones)
return (train_loader, train_sampler, train_logger, train_batch_logger,
optimizer, scheduler)
#normalize = get_normalize_method(mean, std, no_mean_norm, no_std_norm)
normalize = Normalize(mean, std)
spatial_transform = [Resize(sample_size)]
if inference_crop == 'center':
spatial_transform.append(CenterCrop(sample_size))
if input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.append(ToTensor())
spatial_transform.extend([ScaleValue(value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(sample_t_stride))
temporal_transform.append(SlidingWindow(sample_duration, inference_stride))
temporal_transform = TemporalCompose(temporal_transform)
# 加载模型
#print('load model begin!')
model = generate_model_resnet(1) # 生成resnet模型
#model = torch.load('./save_200.pth')
checkpoint = torch.load('./save_200.pth', map_location='cpu')
model.load_state_dict(checkpoint['state_dict'])
#print(model)
model.eval() # 固定batchnorm，dropout等，一定要有
model = model.to(device)
#print('load model done!')
count = 0
# 测试单个视频
# fight
def get_val_utils(opt):
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
spatial_transform = [
Resize(opt.sample_size),
CenterCrop(opt.sample_size),
ToTensor()
]
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.extend([ScaleValue(opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
temporal_transform.append(
TemporalEvenCrop(opt.sample_duration, opt.n_val_samples))
temporal_transform = TemporalCompose(temporal_transform)
val_data_checkpoint_path = opt.result_path / Path('val_data_' +
opt.dataset + '.data')
val_collate_checkpoint_path = opt.result_path / Path('val_coll_' +
opt.dataset + '.data')
if os.path.exists(val_data_checkpoint_path) and os.path.exists(
val_collate_checkpoint_path) and opt.save_load_data_checkpoint:
with open(val_data_checkpoint_path, 'rb') as filehandle:
val_data = pickle.load(filehandle)
with open(val_collate_checkpoint_path, 'rb') as filehandle:
collate_fn = pickle.load(filehandle)
else:
val_data, collate_fn = get_validation_data(
opt.video_path, opt.annotation_path, opt.dataset, opt.input_type,
opt.file_type, spatial_transform, temporal_transform)
if opt.save_load_data_checkpoint:
with open(val_data_checkpoint_path, 'wb') as filehandle:
pickle.dump(val_data, filehandle)
with open(val_collate_checkpoint_path, 'wb') as filehandle:
pickle.dump(collate_fn, filehandle)
if opt.distributed:
val_sampler = torch.utils.data.distributed.DistributedSampler(
val_data, shuffle=False)
else:
val_sampler = None
val_loader = torch.utils.data.DataLoader(val_data,
batch_size=(opt.batch_size //
opt.n_val_samples),
shuffle=False,
num_workers=opt.n_threads,
pin_memory=True,
sampler=val_sampler,
worker_init_fn=worker_init_fn,
collate_fn=collate_fn)
if opt.is_master_node:
val_logger = Logger(opt.result_path / 'val.log',
['epoch', 'loss', 'acc'])
else:
val_logger = None
return val_loader, val_logger
def get_train_utils(opt, model_parameters):
assert opt.train_crop in ['random', 'corner', 'center']
spatial_transform = []
if opt.train_crop == 'random':
spatial_transform.append(
RandomResizedCrop(
opt.sample_size, (opt.train_crop_min_scale, 1.0),
(opt.train_crop_min_ratio, 1.0 / opt.train_crop_min_ratio)))
elif opt.train_crop == 'corner':
scales = [1.0]
scale_step = 1 / (2**(1 / 4))
for _ in range(1, 5):
scales.append(scales[-1] * scale_step)
spatial_transform.append(MultiScaleCornerCrop(opt.sample_size, scales))
elif opt.train_crop == 'center':
spatial_transform.append(Resize(opt.sample_size))
spatial_transform.append(CenterCrop(opt.sample_size))
normalize = get_normalize_method(opt.mean, opt.std, opt.no_mean_norm,
opt.no_std_norm)
if not opt.no_hflip:
spatial_transform.append(RandomHorizontalFlip())
spatial_transform.append(ToArray())
if opt.colorjitter:
spatial_transform.append(ColorJitter())
if opt.input_type == 'flow':
spatial_transform.append(PickFirstChannels(n=2))
spatial_transform.append(ScaleValue(opt.value_scale))
spatial_transform.append(normalize)
spatial_transform = Compose(spatial_transform)
assert opt.train_t_crop in ['random', 'center']
temporal_transform = []
if opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(opt.sample_t_stride))
if opt.train_t_crop == 'random':
temporal_transform.append(TemporalRandomCrop(opt.sample_duration))
elif opt.train_t_crop == 'center':
temporal_transform.append(TemporalCenterCrop(opt.sample_duration))
temporal_transform = TemporalCompose(temporal_transform)
train_data = get_training_data(opt.video_path, opt.annotation_path,
opt.dataset, opt.input_type, opt.file_type,
spatial_transform, temporal_transform)
train_loader = paddle.batch(train_data.reader, batch_size=opt.batch_size)
train_logger = Logger(opt.result_path / 'train.log',
['epoch', 'loss', 'acc', 'lr'])
train_batch_logger = Logger(
opt.result_path / 'train_batch.log',
['epoch', 'batch', 'iter', 'loss', 'acc', 'lr'])
assert opt.lr_scheduler in ['plateau', 'multistep']
assert not (opt.lr_scheduler == 'plateau' and opt.no_val)
if opt.lr_scheduler == 'plateau':
scheduler = ReduceLROnPlateau(learning_rate=opt.learning_rate,
mode='min',
patience=opt.plateau_patience)
else:
scheduler = MultiStepDecay(learning_rate=opt.learning_rate,
milestones=opt.multistep_milestones)
optimizer = fluid.optimizer.MomentumOptimizer(
learning_rate=scheduler,
momentum=opt.momentum,
parameter_list=model_parameters,
use_nesterov=opt.nesterov,
regularization=fluid.regularizer.L2Decay(
regularization_coeff=opt.weight_decay))
return (train_loader, train_logger, train_batch_logger, optimizer,
scheduler)
def score(self):
normalize = get_normalize_method(self.opt.mean, self.opt.std, self.opt.no_mean_norm,
self.opt.no_std_norm)
spatial_transform = [
Resize(self.opt.sample_size),
CenterCrop(self.opt.sample_size),
ToTensor()
]
spatial_transform.extend([ScaleValue(self.opt.value_scale), normalize])
spatial_transform = Compose(spatial_transform)
temporal_transform = []
if self.opt.sample_t_stride > 1:
temporal_transform.append(TemporalSubsampling(self.opt.sample_t_stride))
temporal_transform.append(
TemporalEvenCrop(self.opt.sample_duration, self.opt.n_val_samples))
temporal_transform = TemporalCompose(temporal_transform)
frame_count = get_n_frames(self.opt.video_jpgs_dir_path)
frame_indices = list(range(0, frame_count))
frame_indices = temporal_transform(frame_indices)
spatial_transform.randomize_parameters()
image_name_formatter = lambda x: f'image_{x:05d}.jpg'
loader = VideoLoader(image_name_formatter)
print('frame_indices', frame_indices)
#clips = []
video_outputs = []
model = generate_model(self.opt)
model = load_pretrained_model(model, self.opt.pretrain_path, self.opt.model,
self.opt.n_finetune_classes)
i =0
for frame_indice in frame_indices:
print("%d indice: %s" % (i, str(frame_indice)))
i+=1
clip = loader(self.opt.video_jpgs_dir_path, frame_indice)
clip = [spatial_transform(img) for img in clip]
clip = torch.stack(clip, 0).permute(1, 0, 2, 3)
#parameters = get_fine_tuning_parameters(model, opt.ft_begin_module)
#print('clips:', clips)
#for clip in clips:
with torch.no_grad():
print(clip.shape)
output = model(torch.unsqueeze(clip, 0))
output = F.softmax(output, dim=1).cpu()
#print(output)
video_outputs.append(output[0])
del clip
video_outputs = torch.stack(video_outputs)
average_scores = torch.mean(video_outputs, dim=0)
#inference_loader, inference_class_names = main.get_inference_utils(self.opt)
with self.opt.annotation_path.open('r') as f:
data = json.load(f)
class_to_idx = get_class_labels(data)
idx_to_class = {}
for name, label in class_to_idx.items():
idx_to_class[label] = name
print(idx_to_class)
inference_result = inference.get_video_results(
average_scores, idx_to_class, self.opt.output_topk)
print(inference_result)