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(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.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_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
#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