🛠 NaN fix for PaDim. With custom dataset and backbone distances somet…

…imes contains small negative numbers, possible due to imperfect pseudo inverse calculation. (#392)

anomalib/models/padim/anomaly_map.py

@@ -58,7 +58,7 @@ def compute_distance(embedding: Tensor, stats: List[Tensor]) -> Tensor:
 
         distances = (torch.matmul(delta, inv_covariance) * delta).sum(2).permute(1, 0)
         distances = distances.reshape(batch, height, width)
-        distances = torch.sqrt(distances)
+        distances = distances.clamp(0).sqrt()
 
         return distances