StructuralSimilarityIndexMeasure returns NaN for ([1, 1, 8, 8]) tensor

[anatolicvs]

`
import torch
from torchmetrics import StructuralSimilarityIndexMeasure

preds = torch.rand([1, 1, 8, 8],dtype=torch.float32)
target = preds * 0.75

ssim = StructuralSimilarityIndexMeasure(kernel_size=11, data_range=1.0, reduction='none', sigma=1.5)
ssim = ssim(preds, target)
`
Why the ssim returns NaN for ([1, 1, 8, 8]) tensor ?