Use softmax in _get_gaussian_kernel1d

Author: vfdev-5

torchvision/transforms/functional_tensor.py

@@ -730,8 +730,7 @@ def _get_gaussian_kernel1d(kernel_size: int, sigma: float) -> Tensor:
     ksize_half = (kernel_size - 1) * 0.5
 
     x = torch.linspace(-ksize_half, ksize_half, steps=kernel_size)
-    pdf = torch.exp(-0.5 * (x / sigma).pow(2))
-    kernel1d = pdf / pdf.sum()
+    kernel1d = torch.softmax(-0.5 * (x / sigma).pow(2), dim=0)
 
     return kernel1d