fix(metrics): Corrected dimension count in MS-SSIM error message for 3D images.

[akshatsinha0]

Fixes # .

Description

A few sentences describing the changes proposed in this pull request.

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[coderabbitai]

📝 Walkthrough

Walkthrough

Corrected a dimensionality validation error message in the compute_ms_ssim function within monai/metrics/regression.py. The message now accurately reflects that when spatial_dims is 3, the expected input tensor dimensionality is 5 dimensions rather than 4. This is a single-line fix with no changes to function logic or control flow.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Description check	❓ Inconclusive	The description uses the template structure but the 'Description' section contains only placeholder text ('A few sentences describing...') and lacks actual details about the change.	Replace placeholder text in the Description section with concrete details about why the dimension count was incorrect and what the fix addresses.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the specific change: correcting a dimension count error message in MS-SSIM for 3D images.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[akshatsinha0]

@rijobro

[ericspod]

Looks good to me, thanks!

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

monai/metrics/regression.py (1)

616-638: ⚠️ Potential issue | 🟡 Minor

ssim may be unbound if weights is empty.

ssim is assigned only inside the loop body (line 618). If a caller passes weights=(), len(weights_tensor) == 0, the loop never runs, and ssim at line 636 raises UnboundLocalError.

🛡️ Proposed guard

+    if len(weights_tensor) == 0:
+        raise ValueError("`weights` must not be empty.")
+
     multiscale_list: list[torch.Tensor] = []
     for _ in range(len(weights_tensor)):

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@monai/metrics/regression.py` around lines 616 - 638, The loop uses ssim set
inside compute_ssim_and_cs, so if weights_tensor is empty ssim is unbound; add a
guard before the multiscale loop: if len(weights_tensor) == 0 call
compute_ssim_and_cs(y_pred, y, spatial_dims, data_range, kernel_type,
kernel_size, kernel_sigma, k1, k2) once to produce ssim (and cs), compute the
same reductions (ssim.view(...).mean(1) and relu on that value) and set
multiscale_list accordingly (then stack to multiscale_list_tensor) or
alternatively raise a clear ValueError; update references to ssim,
multiscale_list, compute_ssim_and_cs, avg_pool and multiscale_list_tensor so
code never accesses ssim when weights_tensor is empty.

🧹 Nitpick comments (2)

monai/metrics/regression.py (2)

566-582: compute_ms_ssim docstring is missing a Returns: section.

Per coding guidelines, docstrings must describe return values. The function returns ms_ssim_per_batch but the docstring has no Returns: block.

♻️ Proposed addition

     Raises:
         ValueError: when `y_pred` is not a 2D or 3D image.
+    Returns:
+        ms_ssim_per_batch: MS-SSIM scores per batch element, shape ``(B, 1)``.
     """

As per coding guidelines, "**/*.py: Docstrings should be present for all definitions which describe each variable, return value, and raised exception in the appropriate section of the Google-style of docstrings."

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@monai/metrics/regression.py` around lines 566 - 582, The compute_ms_ssim
docstring is missing a Returns section; add a "Returns:" block describing the
returned value ms_ssim_per_batch (its type, shape and meaning) and any
conditions (e.g., per-batch MS-SSIM scores as a tensor of shape [B] or [B, C]
depending on implementation), and note the unit/range (e.g., values in [0,1]) so
callers know what to expect; update the compute_ms_ssim docstring accordingly to
reference ms_ssim_per_batch and include the return type and brief description.

---

249-253: compute_mean_error_metrics has no docstring.

Per coding guidelines, all definitions require docstrings.

♻️ Proposed docstring

 def compute_mean_error_metrics(y_pred: torch.Tensor, y: torch.Tensor, func: Callable) -> torch.Tensor:
+    """Compute a per-batch mean error metric.
+
+    Args:
+        y_pred: predicted tensor, shape ``(B, C, ...)``.
+        y: ground truth tensor, same shape as ``y_pred``.
+        func: element-wise error function applied to ``(y - y_pred)``.
+
+    Returns:
+        Mean error value per batch element, shape ``(B, 1)``.
+    """
     # reducing in only channel + spatial dimensions (not batch)

As per coding guidelines, "**/*.py: Docstrings should be present for all definitions which describe each variable, return value, and raised exception in the appropriate section of the Google-style of docstrings."

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@monai/metrics/regression.py` around lines 249 - 253, Add a Google-style
docstring to compute_mean_error_metrics describing purpose, parameters, return
value, and any exceptions: explain that the function computes per-sample mean of
a provided error function applied to (y - y_pred), that y_pred and y are
torch.Tensors with batch as dim0 and channel+spatial dims flattened via flt
(partial(torch.flatten, start_dim=1)), and func is a Callable applied
elementwise; document the returned torch.Tensor shape (per-batch mean with
keepdim=True), units/semantics, and note any assumptions or raised errors (e.g.,
mismatched shapes or non-tensor inputs) so callers know preconditions.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@monai/metrics/regression.py`:
- Around line 590-594: Rename the three test methods so unittest discovers and
runs them: change input_ill_input_shape2d to test_input_ill_input_shape2d,
input_ill_input_shape3d to test_input_ill_input_shape3d, and small_inputs to
test_small_inputs; ensure test_input_ill_input_shape2d constructs a 4D y_pred
and calls the metric with spatial_dims=3 so the ValueError path in regression.py
(the block checking "if spatial_dims == 3 and dims != 5") is exercised and
assert the raised error message matches the expected text.

---

Outside diff comments:
In `@monai/metrics/regression.py`:
- Around line 616-638: The loop uses ssim set inside compute_ssim_and_cs, so if
weights_tensor is empty ssim is unbound; add a guard before the multiscale loop:
if len(weights_tensor) == 0 call compute_ssim_and_cs(y_pred, y, spatial_dims,
data_range, kernel_type, kernel_size, kernel_sigma, k1, k2) once to produce ssim
(and cs), compute the same reductions (ssim.view(...).mean(1) and relu on that
value) and set multiscale_list accordingly (then stack to
multiscale_list_tensor) or alternatively raise a clear ValueError; update
references to ssim, multiscale_list, compute_ssim_and_cs, avg_pool and
multiscale_list_tensor so code never accesses ssim when weights_tensor is empty.

---

Nitpick comments:
In `@monai/metrics/regression.py`:
- Around line 566-582: The compute_ms_ssim docstring is missing a Returns
section; add a "Returns:" block describing the returned value ms_ssim_per_batch
(its type, shape and meaning) and any conditions (e.g., per-batch MS-SSIM scores
as a tensor of shape [B] or [B, C] depending on implementation), and note the
unit/range (e.g., values in [0,1]) so callers know what to expect; update the
compute_ms_ssim docstring accordingly to reference ms_ssim_per_batch and include
the return type and brief description.
- Around line 249-253: Add a Google-style docstring to
compute_mean_error_metrics describing purpose, parameters, return value, and any
exceptions: explain that the function computes per-sample mean of a provided
error function applied to (y - y_pred), that y_pred and y are torch.Tensors with
batch as dim0 and channel+spatial dims flattened via flt (partial(torch.flatten,
start_dim=1)), and func is a Callable applied elementwise; document the returned
torch.Tensor shape (per-batch mean with keepdim=True), units/semantics, and note
any assumptions or raised errors (e.g., mismatched shapes or non-tensor inputs)
so callers know preconditions.