Merge branch 'bugfix/error_cosine_sim' of https://github.com/Lightnin…

…g-AI/torchmetrics into bugfix/error_cosine_sim

CHANGELOG.md

@@ -36,6 +36,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Changed minimum supported Pytorch version from 1.8 to 1.10 ([#2145](https://github.com/Lightning-AI/torchmetrics/pull/2145))
 
 
+- Use arange and repeat for deterministic bincount ([#2184](https://github.com/Lightning-AI/torchmetrics/pull/2184))
+
+
 ### Deprecated
 
 - Deprecated `metric._update_called` ([#2141](https://github.com/Lightning-AI/torchmetrics/pull/2141))
@@ -46,7 +49,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Removed
 
--
+- Removed `lpips` third-party package as dependency of `LearnedPerceptualImagePatchSimilarity` metric ([#2230](https://github.com/Lightning-AI/torchmetrics/pull/2230))
 
 
 ### Fixed

requirements/image.txt

@@ -4,4 +4,3 @@
 scipy >1.0.0, <1.11.0
 torchvision >=0.8, <0.17.0
 torch-fidelity <=0.4.0  # bumping to allow install version from master, now used in testing
-lpips <=0.1.4

requirements/image_test.txt

@@ -7,3 +7,4 @@ pytorch-msssim ==1.0.0
 sewar >=0.4.4, <=0.4.6
 numpy <1.25.0
 torch-fidelity @ git+https://github.com/toshas/torch-fidelity@master
+lpips <=0.1.4

src/torchmetrics/image/lpip.py

@@ -47,8 +47,8 @@ class LearnedPerceptualImagePatchSimilarity(Metric):
     Both input image patches are expected to have shape ``(N, 3, H, W)``. The minimum size of `H, W` depends on the
     chosen backbone (see `net_type` arg).
 
-    .. note:: using this metrics requires you to have ``lpips`` package installed. Either install
-        as ``pip install torchmetrics[image]`` or ``pip install lpips``
+    .. note:: using this metrics requires you to have ``torchvision`` package installed. Either install as
+        ``pip install torchmetrics[image]`` or ``pip install torchvision``.
 
     .. note:: this metric is not scriptable when using ``torch<1.8``. Please update your pytorch installation
         if this is a issue.
@@ -71,7 +71,7 @@ class LearnedPerceptualImagePatchSimilarity(Metric):
 
     Raises:
         ModuleNotFoundError:
-            If ``lpips`` package is not installed
+            If ``torchvision`` package is not installed
         ValueError:
             If ``net_type`` is not one of ``"vgg"``, ``"alex"`` or ``"squeeze"``
         ValueError:

src/torchmetrics/utilities/data.py

@@ -169,12 +169,10 @@ def _squeeze_if_scalar(data: Any) -> Any:
 def _bincount(x: Tensor, minlength: Optional[int] = None) -> Tensor:
     """Implement custom bincount.
 
-    PyTorch currently does not support ``torch.bincount`` for:
-
-        - deterministic mode on GPU.
-        - MPS devices
-
-    This implementation fallback to a for-loop counting occurrences in that case.
+    PyTorch currently does not support ``torch.bincount`` when running in deterministic mode on GPU or when running
+    MPS devices or when running on XLA device. This implementation therefore falls back to using a combination of
+    `torch.arange` and `torch.eq` in these scenarios. A small performance hit can expected and higher memory consumption
+    as `[batch_size, mincount]` tensor needs to be initialized compared to native ``torch.bincount``.
 
     Args:
         x: tensor to count
@@ -191,11 +189,11 @@ def _bincount(x: Tensor, minlength: Optional[int] = None) -> Tensor:
     """
     if minlength is None:
         minlength = len(torch.unique(x))
+
     if torch.are_deterministic_algorithms_enabled() or _XLA_AVAILABLE or _TORCH_GREATER_EQUAL_1_12 and x.is_mps:
-        output = torch.zeros(minlength, device=x.device, dtype=torch.long)
-        for i in range(minlength):
-            output[i] = (x == i).sum()
-        return output
+        mesh = torch.arange(minlength, device=x.device).repeat(len(x), 1)
+        return torch.eq(x.reshape(-1, 1), mesh).sum(dim=0)
+
     return torch.bincount(x, minlength=minlength)
 
 

tests/unittests/image/test_lpips.py

@@ -20,7 +20,7 @@
 from torch import Tensor
 from torchmetrics.functional.image.lpips import learned_perceptual_image_patch_similarity
 from torchmetrics.image.lpip import LearnedPerceptualImagePatchSimilarity
-from torchmetrics.utilities.imports import _LPIPS_AVAILABLE
+from torchmetrics.utilities.imports import _LPIPS_AVAILABLE, _TORCHVISION_AVAILABLE
 
 from unittests.helpers import seed_all
 from unittests.helpers.testers import MetricTester
@@ -48,6 +48,7 @@ def _compare_fn(img1: Tensor, img2: Tensor, net_type: str, normalize: bool = Fal
     return res.sum()
 
 
+@pytest.mark.skipif(not _TORCHVISION_AVAILABLE, reason="test requires that torchvision is installed")
 @pytest.mark.skipif(not _LPIPS_AVAILABLE, reason="test requires that lpips is installed")
 class TestLPIPS(MetricTester):
     """Test class for `LearnedPerceptualImagePatchSimilarity` metric."""
@@ -105,6 +106,7 @@ def test_normalize_arg(normalize):
     assert res == res2
 
 
+@pytest.mark.skipif(not _TORCHVISION_AVAILABLE, reason="test requires that torchvision is installed")
 @pytest.mark.skipif(not _LPIPS_AVAILABLE, reason="test requires that lpips is installed")
 def test_error_on_wrong_init():
     """Test class raises the expected errors."""
@@ -115,6 +117,7 @@ def test_error_on_wrong_init():
         LearnedPerceptualImagePatchSimilarity(net_type="squeeze", reduction=None)
 
 
+@pytest.mark.skipif(not _TORCHVISION_AVAILABLE, reason="test requires that torchvision is installed")
 @pytest.mark.skipif(not _LPIPS_AVAILABLE, reason="test requires that lpips is installed")
 @pytest.mark.parametrize(
     ("inp1", "inp2"),