Change semantic ordering of kernel_size parameter in SSIM

CHANGELOG.md

@@ -36,6 +36,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - Fixed bug where compositional metrics where unable to sync because of type mismatch ([#454](https://github.com/PyTorchLightning/metrics/pull/454))
 
+
+- Fixed the semantic ordering of kernel height and width in `SSIM` metric ([#474](https://github.com/PyTorchLightning/metrics/pull/474))
+
+
 ## [0.5.0] - 2021-08-09
 
 ### Added

tests/image/test_ssim.py

@@ -44,7 +44,7 @@
  )
 
 
-def _sk_ssim(preds, target, data_range, multichannel):
+def _sk_ssim(preds, target, data_range, multichannel, kernel_size):
  c, h, w = preds.shape[-3:]
  sk_preds = preds.view(-1, c, h, w).permute(0, 2, 3, 1).numpy()
  sk_target = target.view(-1, c, h, w).permute(0, 2, 3, 1).numpy()
@@ -58,7 +58,7 @@ def _sk_ssim(preds, target, data_range, multichannel):
  data_range=data_range,
  multichannel=multichannel,
  gaussian_weights=True,
- win_size=11,
+ win_size=kernel_size,
  sigma=1.5,
  use_sample_covariance=False,
  )
@@ -68,38 +68,39 @@ def _sk_ssim(preds, target, data_range, multichannel):
  "preds, target, multichannel",
  [(i.preds, i.target, i.multichannel) for i in _inputs],
 )
+@pytest.mark.parametrize("kernel_size", [5, 11])
 class TestSSIM(MetricTester):
- atol = 6e-5
+ atol = 6e-3
 
  @pytest.mark.parametrize("ddp", [True, False])
  @pytest.mark.parametrize("dist_sync_on_step", [True, False])
- def test_ssim(self, preds, target, multichannel, ddp, dist_sync_on_step):
+ def test_ssim(self, preds, target, multichannel, kernel_size, ddp, dist_sync_on_step):
  self.run_class_metric_test(
  ddp,
  preds,
  target,
  SSIM,
- partial(_sk_ssim, data_range=1.0, multichannel=multichannel),
- metric_args={"data_range": 1.0},
+ partial(_sk_ssim, data_range=1.0, multichannel=multichannel, kernel_size=kernel_size),
+ metric_args={"data_range": 1.0, "kernel_size": (kernel_size, kernel_size)},
  dist_sync_on_step=dist_sync_on_step,
  )
 
- def test_ssim_functional(self, preds, target, multichannel):
+ def test_ssim_functional(self, preds, target, multichannel, kernel_size):
  self.run_functional_metric_test(
  preds,
  target,
  ssim,
- partial(_sk_ssim, data_range=1.0, multichannel=multichannel),
- metric_args={"data_range": 1.0},
+ partial(_sk_ssim, data_range=1.0, multichannel=multichannel, kernel_size=kernel_size),
+ metric_args={"data_range": 1.0, "kernel_size": (kernel_size, kernel_size)},
  )
 
  # SSIM half + cpu does not work due to missing support in torch.log
  @pytest.mark.xfail(reason="SSIM metric does not support cpu + half precision")
- def test_ssim_half_cpu(self, preds, target, multichannel):
+ def test_ssim_half_cpu(self, preds, target, multichannel, kernel_size):
  self.run_precision_test_cpu(preds, target, SSIM, ssim, {"data_range": 1.0})
 
  @pytest.mark.skipif(not torch.cuda.is_available(), reason="test requires cuda")
- def test_ssim_half_gpu(self, preds, target, multichannel):
+ def test_ssim_half_gpu(self, preds, target, multichannel, kernel_size):
  self.run_precision_test_gpu(preds, target, SSIM, ssim, {"data_range": 1.0})
 
 
@@ -127,3 +128,40 @@ def test_ssim_invalid_inputs(pred, target, kernel, sigma):
  target = torch.rand(target)
  with pytest.raises(ValueError):
  ssim(pred, target, kernel, sigma)
+
+
+def test_ssim_unequal_kernel_size():
+ """Test the case where kernel_size[0] != kernel_size[1]"""
+ preds = torch.tensor(
+ [
+ [
+ [
+ [1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0],
+ [1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0],
+ [1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+ [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0],
+ [0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0],
+ [0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0],
+ [1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0],
+ ]
+ ]
+ ]
+ )
+ target = torch.tensor(
+ [
+ [
+ [
+ [1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0],
+ [0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0],
+ [1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0],
+ [1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0],
+ [1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0],
+ [0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0],
+ [0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0],
+ ]
+ ]
+ ]
+ )
+ # kernel order matters
+ assert ssim(preds, target, kernel_size=(3, 5)) == torch.tensor(0.10814697)
+ assert ssim(preds, target, kernel_size=(5, 3)) != torch.tensor(0.10814697)

torchmetrics/functional/image/ssim.py

@@ -146,11 +146,11 @@ def _ssim_compute(
  channel = preds.size(1)
  dtype = preds.dtype
  kernel = _gaussian_kernel(channel, kernel_size, sigma, dtype, device)
- pad_w = (kernel_size[0] - 1) // 2
- pad_h = (kernel_size[1] - 1) // 2
+ pad_h = (kernel_size[0] - 1) // 2
+ pad_w = (kernel_size[1] - 1) // 2
 
- preds = F.pad(preds, (pad_w, pad_w, pad_h, pad_h), mode="reflect")
- target = F.pad(target, (pad_w, pad_w, pad_h, pad_h), mode="reflect")
+ preds = F.pad(preds, (pad_h, pad_h, pad_w, pad_w), mode="reflect")
+ target = F.pad(target, (pad_h, pad_h, pad_w, pad_w), mode="reflect")
 
  input_list = torch.cat((preds, target, preds * preds, target * target, preds * target)) # (5 * B, C, H, W)
  outputs = F.conv2d(input_list, kernel, groups=channel)