o oi@spddlmZddlmZddlZddlmmZddlmZddl m Z   ddddZ Gdddej Z dS)) annotations)OptionalN)nn)mask_ignore_pixels:0yE>pred torch.Tensortargetalphafloatbetaeps ignore_index Optional[int]returncCsdt|tjstdt|t|jdkstd|j|jdd|jddks8td|jd|j|j|jksJtd|jd|jt j |d d }t ||\}}| d | d }|dury| d j|jd } || }| d } n|j\} } } }tj| f| ||j|jd } |d }tj|| |t| ||dd|}||}d|S)uCriterion that computes Tversky Coefficient loss. According to :cite:`salehi2017tversky`, we compute the Tversky Coefficient as follows: .. math:: \text{S}(P, G, \alpha; \beta) = \frac{|PG|}{|PG| + \alpha |P \setminus G| + \beta |G \setminus P|} Where: - :math:`P` and :math:`G` are the predicted and ground truth binary labels. - :math:`\alpha` and :math:`\beta` control the magnitude of the penalties for FPs and FNs, respectively. Note: - :math:`\alpha = \beta = 0.5` => dice coeff - :math:`\alpha = \beta = 1` => tanimoto coeff - :math:`\alpha + \beta = 1` => F beta coeff Args: pred: logits tensor with shape :math:`(N, C, H, W)` where C = number of classes. target: labels tensor with shape :math:`(N, H, W)` where each value is :math:`0 ≤ targets[i] ≤ C-1`. alpha: the first coefficient in the denominator. beta: the second coefficient in the denominator. eps: scalar for numerical stability. ignore_index: labels with this value are ignored in the loss computation. Return: the computed loss. Example: >>> N = 5 # num_classes >>> pred = torch.randn(1, N, 3, 5, requires_grad=True) >>> target = torch.empty(1, 3, 5, dtype=torch.long).random_(N) >>> output = tversky_loss(pred, target, alpha=0.5, beta=0.5) >>> output.backward() z%pred type is not a torch.Tensor. Got z,Invalid pred shape, we expect BxNxHxW. Got: Nz.pred and target shapes must be the same. Got: z and z1pred and target must be in the same device. Got: )dim)dtype)r)rdeviceg?)value) isinstancetorchTensor TypeErrortypelenshape ValueErrorrFsoftmaxrgather unsqueezetorsumfulladdcmul full_likeadd_divmean)rr r r rr pred_soft target_maskp_truemtotalB_HW intersection denominatorscorer;I/home/ubuntu/.local/lib/python3.10/site-packages/kornia/losses/tversky.py tversky_loss s8 0     r=cs.eZdZdZddfd d ZdddZZS) TverskyLossuCriterion that computes Tversky Coefficient loss. According to :cite:`salehi2017tversky`, we compute the Tversky Coefficient as follows: .. math:: \text{S}(P, G, \alpha; \beta) = \frac{|PG|}{|PG| + \alpha |P \setminus G| + \beta |G \setminus P|} Where: - :math:`P` and :math:`G` are the predicted and ground truth binary labels. - :math:`\alpha` and :math:`\beta` control the magnitude of the penalties for FPs and FNs, respectively. Note: - :math:`\alpha = \beta = 0.5` => dice coeff - :math:`\alpha = \beta = 1` => tanimoto coeff - :math:`\alpha + \beta = 1` => F beta coeff Args: alpha: the first coefficient in the denominator. beta: the second coefficient in the denominator. eps: scalar for numerical stability. ignore_index: labels with this value are ignored in the loss computation. Shape: - Pred: :math:`(N, C, H, W)` where C = number of classes. - Target: :math:`(N, H, W)` where each value is :math:`0 ≤ targets[i] ≤ C-1`. Examples: >>> N = 5 # num_classes >>> criterion = TverskyLoss(alpha=0.5, beta=0.5) >>> pred = torch.randn(1, N, 3, 5, requires_grad=True) >>> target = torch.empty(1, 3, 5, dtype=torch.long).random_(N) >>> output = criterion(pred, target) >>> output.backward() rrr r r rrrrNonecs&t||_||_||_||_dSN)super__init__r r rr)selfr r rr __class__r;r<rBs  zTverskyLoss.__init__rr r cCst|||j|j|j|jSr@)r=r r rr)rCrr r;r;r<forwardszTverskyLoss.forwardrr) r r r r rr rrrr?)rr r r rr )__name__ __module__ __qualname____doc__rBrF __classcell__r;r;rDr<r>xs)r>rG)rr r r r r r r rr rrrr ) __future__rtypingrrtorch.nn.functionalr functionalr#kornia.losses._utilsrr=Moduler>r;r;r;r<s      X