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_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_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 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)
