Merge branch 'dev' into patch-1

Author: emianaya

CONTRIBUTING.md

@@ -303,6 +303,13 @@ By making a contribution to this project, I certify that:
     this project or the open source license(s) involved.
 ```
 
+> **Tip:** If you need to add a DCO remediation commit (e.g., after a force-push
+> or rebase), include `[skip ci]` in the commit message so the remediation
+> does not trigger unnecessary CI pipelines:
+> ```bash
+> git commit -s --allow-empty -m 'DCO Remediation Commit for... [skip ci]'
+> ```
+
 #### Utility functions
 
 MONAI provides a set of generic utility functions and frequently used routines.
@@ -358,6 +365,90 @@ Ideally, the new branch should be based on the latest `dev` branch.
 1. Reviewer and contributor may have discussions back and forth until all comments addressed.
 1. Wait for the pull request to be merged.
 
+## Skipping CI
+
+MONAI's CI pipelines run automatically on every push and pull request.
+These pipelines can be resource-intensive, especially the full premerge matrix
+which spans multiple OSes, Python versions, and PyTorch versions.
+
+To reduce unnecessary resource consumption and speed up iteration, you can
+skip CI on commits that don't need automated validation — for example,
+documentation-only changes, README updates, workflow YAML changes, or WIP
+commits during development.
+
+### Mechanism
+
+GitHub Actions natively supports skipping `push` and `pull_request` workflows
+when the commit message contains any of the following strings:
+
+- `[skip ci]`
+- `[ci skip]`
+- `[no ci]`
+- `[skip actions]`
+- `[actions skip]`
+
+These are case-insensitive. `[skip ci]` is the recommended convention for
+this repository.
+
+Alternatively, you can add a `skip-checks: true` trailer at the end of the
+commit message, preceded by two blank lines:
+
+```
+commit message
+
+skip-checks: true
+```
+
+### Usage
+
+Add the keyword anywhere in the commit message when committing:
+
+```bash
+git commit -s -m 'update docs [skip ci]'
+```
+
+If the HEAD commit of a pull request contains the skip instruction,
+the entire PR's pull_request-triggered workflows are skipped.
+
+### Which workflows are affected
+
+The skip instruction applies only to workflows triggered by `on: push` or
+`on: pull_request` events. All other workflows — those using `issue_comment`,
+`repository_dispatch`, `schedule`, or `workflow_dispatch` — use different
+event types and are **not** affected by `[skip ci]`.
+
+### Important caveat
+
+If a workflow is skipped via `[skip ci]`, its associated checks remain in
+"Pending" state. If your pull request requires those checks to pass before
+merging, you will need to push a new commit **without** the skip instruction
+to trigger the CI pipelines.
+
+### When to use
+
+Use `[skip ci]` for commits that are safe to skip CI:
+
+- Documentation-only changes (`docs/`, `README.md`, docstrings)
+- Workflow configuration changes (`.github/`)
+- Repository metadata (`.gitignore`, `CONTRIBUTING.md`, `LICENSE`)
+- WIP or draft commits during local development
+
+Do **not** use `[skip ci]` for commits that change:
+
+- Source code in `monai/`
+- Test files in `tests/`
+- Dependencies (`requirements*.txt`, `setup.cfg`, `setup.py`)
+- Anything that could affect correctness or compatibility
+
+### Quick example
+
+```bash
+git commit -s -m 'fix typo in README [skip ci]'
+```
+
+This commit will be recorded in the repository history but will not
+consume CI minutes.
+
 ## The code reviewing process
 
 ### Reviewing pull requests

docs/source/lazy_resampling.rst

@@ -253,7 +253,7 @@ so the user must set lazy=True on the transforms that they still wish to execute
 
 .. figure:: ../images/lazy_resampling_none_example.svg
 
-    Figure shwoing the effect of using ``lazy=False`` when ``Compose`` is being executed with ``lazy=None``. Note that
+    Figure showing the effect of using ``lazy=False`` when ``Compose`` is being executed with ``lazy=None``. Note that
     the additional resamples that occur due to ``RandRotate90d`` being executed in a non-lazy fashion.
 
 
@@ -270,4 +270,4 @@ the following transform is a lazy transform, or is configured to execute lazily.
 .. figure:: ../images/lazy_resampling_apply_pending_example.svg
 
     Figure showing the use of :class:`ApplyPendingd<monai.transforms.lazy.dictionary.ApplyPendingd>` to cause
-    resampling to occur in the midele of a chain of lazy transforms.
+    resampling to occur in the middle of a chain of lazy transforms.

docs/source/modules.md

@@ -205,7 +205,7 @@ The workflow and some of MONAI event handlers are shown as below [[Workflow exam
 
 ### EnsembleEvaluator
 
-A typical ensemble procoess is implemented as a ready-to-use workflow [[Cross validation and model ensemble tutorial]](https://github.com/Project-MONAI/tutorials/blob/main/modules/cross_validation_models_ensemble.ipynb):
+A typical ensemble process is implemented as a ready-to-use workflow [[Cross validation and model ensemble tutorial]](https://github.com/Project-MONAI/tutorials/blob/main/modules/cross_validation_models_ensemble.ipynb):
 1. Split all the training dataset into K folds.
 2. Train K models with every K-1 folds data.
 3. Execute inference on the test data with all the K models.

docs/source/whatsnew_1_5.md

@@ -3,7 +3,7 @@
 
 - Support numpy 2.x and Pytorch 2.6
 - MAISI inference accelerate
-- Bundles storage changed to huggingface and correspoinding api updated in core
+- Bundles storage changed to huggingface and corresponding api updated in core
 - Ported remaining generative tutorials and bundles
 - New tutorials:
   - [2d_regression/image_restoration.ipynb](https://github.com/Project-MONAI/tutorials/blob/main/2d_regression/image_restoration.ipynb)

monai/data/meta_tensor.py

@@ -495,7 +495,8 @@ def astype(self, dtype, device=None, *_args, **_kwargs):
             _kwargs: additional kwargs (currently unused).
 
         Returns:
-            data array instance
+            ``MetaTensor`` when a torch dtype is given (metadata is preserved),
+            or ``np.ndarray`` when a numpy dtype is given.
         """
         if isinstance(dtype, str):
             mod_str, *dtype = dtype.split(".", 1)
@@ -506,7 +507,7 @@ def astype(self, dtype, device=None, *_args, **_kwargs):
 
         out_type: type[torch.Tensor] | type[np.ndarray] | None
         if mod_str == "torch":
-            out_type = torch.Tensor
+            out_type = type(self)
         elif mod_str in ("numpy", "np"):
             out_type = np.ndarray
         else:

monai/handlers/utils.py

@@ -122,15 +122,15 @@ class    mean    median    max    5percentile 95percentile  notnans
 
             # add the average value of all classes to v
             if class_labels is None:
-                class_labels = ["class" + str(i) for i in range(v.shape[1])]
+                labels = ["class" + str(i) for i in range(v.shape[1])]
             else:
-                class_labels = [str(i) for i in class_labels]  # ensure to have a list of str
+                labels = [str(i) for i in class_labels]  # ensure to have a list of str
 
-            class_labels += ["mean"]
+            labels += ["mean"]
             v = np.concatenate([v, np.nanmean(v, axis=1, keepdims=True)], axis=1)
 
             with open(os.path.join(save_dir, f"{k}_raw.csv"), "w") as f:
-                f.write(f"filename{deli}{deli.join(class_labels)}\n")
+                f.write(f"filename{deli}{deli.join(labels)}\n")
                 for i, b in enumerate(v):
                     f.write(
                         f"{images[i] if images is not None else str(i)}{deli}"
@@ -164,7 +164,7 @@ def _compute_op(op: str, d: np.ndarray) -> Any:
                 with open(os.path.join(save_dir, f"{k}_summary.csv"), "w") as f:
                     f.write(f"class{deli}{deli.join(ops)}\n")
                     for i, c in enumerate(np.transpose(v)):
-                        f.write(f"{class_labels[i]}{deli}{deli.join([f'{_compute_op(k, c):.4f}' for k in ops])}\n")
+                        f.write(f"{labels[i]}{deli}{deli.join([f'{_compute_op(k, c):.4f}' for k in ops])}\n")
 
 
 def from_engine(keys: KeysCollection, first: bool = False) -> Callable:

monai/transforms/utils.py

@@ -1224,15 +1224,15 @@ def get_largest_connected_component_mask(
     if num_features <= num_components:
         out = img_.astype(bool)
     else:
-        # ignore background
-        nonzeros = features[lib.nonzero(features)]
-        # get number voxels per feature (bincount). argsort[::-1] to get indices
-        # of largest components.
-        features_to_keep = lib.argsort(lib.bincount(nonzeros))[::-1]
-        # only keep the first n non-background indices
-        features_to_keep = features_to_keep[:num_components]
-        # generate labelfield. True if in list of features to keep
-        out = lib.isin(features, features_to_keep)
+        # bincount counts every label; index 0 is background, so drop it before ranking
+        counts = lib.bincount(features.reshape(-1))
+        counts[0] = 0
+        # argsort[::-1] gives labels of the largest components; keep the first n
+        features_to_keep = lib.argsort(counts)[::-1][:num_components]
+        # boolean lookup-table gather over the label field, cheaper than isin
+        keep = lib.zeros(counts.shape[0], dtype=bool)
+        keep[features_to_keep] = True
+        out = keep[features]
 
     return convert_to_dst_type(out, dst=img, dtype=out.dtype)[0]
 

tests/data/meta_tensor/test_meta_tensor.py

@@ -435,8 +435,12 @@ def test_astype(self):
         for np_types in ("float32", "np.float32", "numpy.float32", np.float32, float, "int", np.uint16):
             self.assertIsInstance(t.astype(np_types), np.ndarray)
         for pt_types in ("torch.float", torch.float, "torch.float64"):
-            self.assertIsInstance(t.astype(pt_types), torch.Tensor)
-        self.assertIsInstance(t.astype("torch.float", device="cpu"), torch.Tensor)
+            result = t.astype(pt_types)
+            self.assertIsInstance(result, MetaTensor)
+            self.assertEqual(result.meta.get("fname"), "filename")
+        result = t.astype("torch.float", device="cpu")
+        self.assertIsInstance(result, MetaTensor)
+        self.assertEqual(result.meta.get("fname"), "filename")
 
     def test_transforms(self):
         key = "im"

tests/handlers/test_write_metrics_reports.py

@@ -63,6 +63,28 @@ def test_content(self):
             self.assertTrue(os.path.exists(os.path.join(tempdir, "metric4_raw.csv")))
             self.assertTrue(os.path.exists(os.path.join(tempdir, "metric4_summary.csv")))
 
+    def test_multi_metric_details_headers(self):
+        with tempfile.TemporaryDirectory() as tempdir:
+            write_metrics_reports(
+                save_dir=Path(tempdir),
+                images=["img1", "img2"],
+                metrics=None,
+                metric_details={
+                    "m1": torch.tensor([[1, 2, 3], [4, 5, 6]]),
+                    "m2": torch.tensor([[7, 8], [9, 10]]),
+                    "m3": torch.tensor([[11, 12, 13, 14], [15, 16, 17, 18]]),
+                },
+                summary_ops=None,
+                deli=",",
+                output_type="csv",
+            )
+            for name, nclass in [("m1", 3), ("m2", 2), ("m3", 4)]:
+                path = os.path.join(tempdir, f"{name}_raw.csv")
+                self.assertTrue(os.path.exists(path))
+                with open(path) as f:
+                    header = f.readline().strip().split(",")
+                self.assertEqual(header, ["filename"] + [f"class{i}" for i in range(nclass)] + ["mean"])
+
 
 if __name__ == "__main__":
     unittest.main()