def test_multiclass_lovasz():
inp = torch.randn(BS, N_CLASSES, IM_SIZE, IM_SIZE).float()
target = torch.randint(0, N_CLASSES, (BS, 1, IM_SIZE, IM_SIZE)).float()
criterion = losses.LovaszLoss(mode="multiclass")
loss = criterion(inp, target)
# try other target shape
target = target.view(BS, IM_SIZE, IM_SIZE)
loss2 = criterion(inp, target)
assert torch.allclose(loss, loss2)
def test_binary_lovasz():
inp = torch.randn(BS, 1, IM_SIZE, IM_SIZE).float()
target = torch.randint(0, 2, (BS, 1, IM_SIZE, IM_SIZE)).float()
criterion = losses.LovaszLoss(mode="binary")
loss = criterion(inp, target)
# try other target shape
target = target.view(BS, IM_SIZE, IM_SIZE)
loss2 = criterion(inp, target)
assert torch.allclose(loss, loss2)
def test_multiclass_multilabel_lovasz():
loss = losses.LovaszLoss(mode="multiclass")(INP_IMG, TARGET_IMG_MULTICLASS)
loss2 = losses.LovaszLoss(mode="multilabel")(INP_IMG,
TARGET_IMG_MULTILABEL)
assert torch.allclose(loss, loss2)
def test_binary_lovasz():
loss = losses.LovaszLoss(mode="binary")(INP_IMG_BINARY, TARGET_IMG_BINARY)
# try other target shape
target = TARGET_IMG_BINARY.view(BS, IM_SIZE, IM_SIZE)
loss2 = losses.LovaszLoss(mode="binary")(INP_IMG_BINARY, target)
assert torch.allclose(loss, loss2)
def test_multilabel_lovasz():
inp = torch.randn(BS, N_CLASSES, IM_SIZE, IM_SIZE).float()
target = torch.randint(0, 2, (BS, N_CLASSES, IM_SIZE, IM_SIZE)).float()
criterion = losses.LovaszLoss(mode="multilabel")
loss = criterion(inp, target)
assert loss