def test():
from train_config import get_config
from tools.utils_for_weakly import get_bbox_instance
args = get_config()
data_loader = get_cityscapes_dataloader(args, True)
for i, sample in enumerate(data_loader):
image = sample['image']
instance = sample['instance']
label = sample['label']
# bbox = sample['bbox']
bbox_instance = get_bbox_instance(instance[0], label[0])
bboxes = torch.zeros(1, image.size(-2), image.size(-1))
for i in bbox_instance:
box_temp = i['mask'].unsqueeze(dim=0)
bboxes[box_temp > 0] = i['cls']
transforms.ToPILImage()(image[0]).show()
transforms.ToPILImage()(bboxes).show()
# transforms.ToPILImage()(instance[0]*10).show()
# transforms.ToPILImage()(label[0]*50).show()
# transforms.ToPILImage()(bbox_instance[0]*20).show()
# bbox_temp = bbox[0].argmax(dim=0).float()
# transforms.ToPILImage()(bbox_temp*50).show()
print(instance.unique())
print(label.unique())
# print(bbox_temp.unique())
print('Next')
# print(sample)
pass
def cat_data_set_test():
from train_config import get_config
args = get_config()
cat_data_set = get_cityscapes_dataset(args, train=True)
print(len(cat_data_set))
data_loader = DataLoader(cat_data_set, shuffle=True)
for sample in data_loader:
print(sample['image'].shape)
pass
def _test_freeze_batch_norm():
from train_config import get_config
from models import get_model
from models.sync_batchnorm import convert_model
args = get_config()
model = get_model(args)
model = convert_model(model)
for m in model._modules:
print(m)
model = nn.DataParallel(model)
freeze_batch_norm(model)
check_batch_norm_freeze(model)
def test_weakly_dataset():
from train_config import get_config
args = get_config()
args.data_choose_size = 914
data_loader_1 = get_cityscapes_dataset(args, True)
args.data_choose_size = -914
data_loader_2 = get_cityscapes_dataset(args, True)
# check overlap:
for img1 in data_loader_1.image_list:
for img2 in data_loader_2.image_list:
if img1 == img2:
raise Exception('Overlap!')
pass
def test_label_data_set():
from train_config import get_config
args = get_config()
data_set = BalancedCityscapesDataset(crop_size=512,
fake_size=41,
label=6,
root_dir=args.cityscapes_data_path,
type="train",
choose_size=914,
transform=None,
repeat=1)
print(data_set.__len__())
for sample in data_set:
image = sample['image']
image.show()