fix(image): correct block_size validation in PSNRB (and → or)

src/torchmetrics/functional/image/psnrb.py

@@ -131,6 +131,9 @@ def peak_signal_noise_ratio_with_blocked_effect(
         tensor(7.8402)
 
     """
+    if not isinstance(block_size, int) or block_size < 1:
+        raise ValueError("Argument ``block_size`` should be a positive integer")
+
     if isinstance(data_range, tuple):
         preds = torch.clamp(preds, min=data_range[0], max=data_range[1])
         target = torch.clamp(target, min=data_range[0], max=data_range[1])

src/torchmetrics/image/psnrb.py

@@ -79,7 +79,7 @@ def __init__(
         **kwargs: Any,
     ) -> None:
         super().__init__(**kwargs)
-        if not isinstance(block_size, int) and block_size < 1:
+        if not isinstance(block_size, int) or block_size < 1:
             raise ValueError("Argument ``block_size`` should be a positive integer")
         self.block_size = block_size
 

tests/unittests/image/test_psnrb.py

@@ -135,5 +135,21 @@ def test_psnr_half_gpu(self, preds, target):
 
 def test_error_on_color_images():
     """Test that appropriate error is raised when color images are passed to PSNRB metric."""
-    with pytest.raises(ValueError, match="`psnrb` metric expects grayscale images.*"):
+    with pytest.raises(ValueError, match=r"`psnrb` metric expects grayscale images.*"):
         peak_signal_noise_ratio_with_blocked_effect(torch.rand(1, 3, 16, 16), torch.rand(1, 3, 16, 16), data_range=1.0)
+
+
+@pytest.mark.parametrize("block_size", [0, -5, 1.5, "foo"])
+def test_error_on_invalid_block_size_class(block_size):
+    """Test that ValueError is raised for non-positive or non-integer block_size in the class API."""
+    with pytest.raises(ValueError, match="Argument ``block_size`` should be a positive integer"):
+        PeakSignalNoiseRatioWithBlockedEffect(data_range=1.0, block_size=block_size)
+
+
+@pytest.mark.parametrize("block_size", [0, -5, 1.5, "foo"])
+def test_error_on_invalid_block_size_functional(block_size):
+    """Test that ValueError is raised for non-positive or non-integer block_size in the functional API."""
+    with pytest.raises(ValueError, match="Argument ``block_size`` should be a positive integer"):
+        peak_signal_noise_ratio_with_blocked_effect(
+            torch.rand(1, 1, 16, 16), torch.rand(1, 1, 16, 16), data_range=1.0, block_size=block_size
+        )