def test_raise(self):
name: str = 'InvalidLossName'
params: dict = {}
with self.assertRaises(ValueError):
cfg_tools.load_loss(name, **params)
def test_jaccard_loss(self):
name: str = 'JaccardLoss'
params: dict = {'num_classes': 5}
criterion = cfg_tools.load_loss(name, **params)
self.assertIsInstance(criterion, losses.JaccardLoss)
def test_cross_entropy(self):
name: str = 'CrossEntropyLoss'
params: dict = {}
criterion = cfg_tools.load_loss(name, **params)
self.assertIsInstance(criterion, nn.CrossEntropyLoss)
def test_focal_loss(self):
name: str = 'FocalLoss'
params: dict = {}
criterion = cfg_tools.load_loss(name, **params)
self.assertIsInstance(criterion, losses.FocalLoss)
cfg_dict: Dict[str, Any] = utils.load_yaml(args.config)
cfg: utils.DotDict = utils.DotDict(cfg_dict)
device: str = 'cuda' if torch.cuda.is_available() else 'cpu'
model = models.utils.load_model(num_classes=cfg.num_classes,
architecture=cfg.model.architecture,
backbone=cfg.model.backbone,
pretrained=True)
model.load_state_dict(torch.load(args.weights_path, map_location=device))
model.eval()
model = model.to(device)
logger.info(f'Configurations: {cfg}')
criterion = cfg_tools.load_loss(cfg.loss.name, **cfg.loss.params)
_, X_test, _, y_test = load_dataset()
dtest = SegmentationDataset(X=X_test,
y=y_test,
num_classes=cfg.num_classes,
img_size=cfg.img_size,
transforms=albu.core.serialization.from_dict(
cfg.albumentations.eval))
test_loader = torch.utils.data.DataLoader(dtest,
batch_size=cfg.batch_size,
shuffle=False,
drop_last=False)
model.eval()