Merge branch 'dev' into fix-6840

Author: KumoLiu

.github/workflows/codeql-analysis.yml

@@ -42,7 +42,7 @@ jobs:
 
     # Initializes the CodeQL tools for scanning.
     - name: Initialize CodeQL
-      uses: github/codeql-action/init@v3
+      uses: github/codeql-action/init@v4
       with:
         languages: ${{ matrix.language }}
         # If you wish to specify custom queries, you can do so here or in a config file.
@@ -72,4 +72,4 @@ jobs:
         BUILD_MONAI=1 ./runtests.sh --build
 
     - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@v3
+      uses: github/codeql-action/analyze@v4

.github/workflows/docker.yml

@@ -37,7 +37,7 @@ jobs:
           python setup.py build
           cat build/lib/monai/_version.py
       - name: Upload version
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v5
         with:
           name: _version.py
           path: build/lib/monai/_version.py
@@ -57,7 +57,7 @@ jobs:
       with:
         ref: dev
     - name: Download version
-      uses: actions/download-artifact@v5
+      uses: actions/download-artifact@v6
       with:
         name: _version.py
     - name: docker_build

.github/workflows/integration.yml

@@ -74,7 +74,7 @@ jobs:
       run: ./runtests.sh --build --net
 
     - name: Add reaction
-      uses: peter-evans/create-or-update-comment@v4
+      uses: peter-evans/create-or-update-comment@v5
       with:
         token: ${{ secrets.PR_MAINTAIN }}
         repository: ${{ github.event.client_payload.github.payload.repository.full_name }}
@@ -154,7 +154,7 @@ jobs:
         python -m tests.test_integration_gpu_customization
 
     - name: Add reaction
-      uses: peter-evans/create-or-update-comment@v4
+      uses: peter-evans/create-or-update-comment@v5
       with:
         token: ${{ secrets.PR_MAINTAIN }}
         repository: ${{ github.event.client_payload.github.payload.repository.full_name }}

.github/workflows/pythonapp-min.yml

@@ -88,6 +88,7 @@ jobs:
       run: |
         which python
         python -m pip install --user --upgrade pip setuptools wheel
+        python -m pip install --user more-itertools>=8.0
     - name: cache weekly timestamp
       id: pip-cache
       run: |
@@ -136,6 +137,7 @@ jobs:
       run: |
         which python
         python -m pip install --user --upgrade pip setuptools wheel
+        python -m pip install --user more-itertools>=8.0
     - name: cache weekly timestamp
       id: pip-cache
       run: |

.github/workflows/release.yml

@@ -66,7 +66,7 @@ jobs:
 
     - if: matrix.python-version == '3.9' && startsWith(github.ref, 'refs/tags/')
       name: Upload artifacts
-      uses: actions/upload-artifact@v4
+      uses: actions/upload-artifact@v5
       with:
         name: dist
         path: dist/
@@ -109,7 +109,7 @@ jobs:
           python setup.py build
           cat build/lib/monai/_version.py
       - name: Upload version
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v5
         with:
           name: _version.py
           path: build/lib/monai/_version.py
@@ -127,7 +127,7 @@ jobs:
     steps:
       - uses: actions/checkout@v5
       - name: Download version
-        uses: actions/download-artifact@v5
+        uses: actions/download-artifact@v6
         with:
           name: _version.py
       - name: Set tag

docs/source/transforms.rst

@@ -37,6 +37,11 @@ Generic Interfaces
 .. autoclass:: MultiSampleTrait
     :members:
 
+`ReduceTrait`
+^^^^^^^^^^^^^^^^^^
+.. autoclass:: ReduceTrait
+    :members:
+
 `Randomizable`
 ^^^^^^^^^^^^^^
 .. autoclass:: Randomizable
@@ -1252,6 +1257,12 @@ Utility
     :members:
     :special-members: __call__
 
+`FlattenSequence`
+""""""""""""""""""""""""
+.. autoclass:: FlattenSequence
+    :members:
+    :special-members: __call__
+
 Dictionary Transforms
 ---------------------
 
@@ -2337,6 +2348,12 @@ Utility (Dict)
     :members:
     :special-members: __call__
 
+`FlattenSequenced`
+"""""""""""""""""""""""""
+.. autoclass:: FlattenSequenced
+    :members:
+    :special-members: __call__
+
 
 MetaTensor
 ^^^^^^^^^^

monai/data/box_utils.py

@@ -826,7 +826,10 @@ def box_iou(boxes1: NdarrayOrTensor, boxes2: NdarrayOrTensor) -> NdarrayOrTensor
         boxes2: bounding boxes, Mx4 or Mx6 torch tensor or ndarray. The box mode is assumed to be ``StandardMode``
 
     Returns:
-        IoU, with size of (N,M) and same data type as ``boxes1``
+        An array/tensor matching the container type of ``boxes1`` (NumPy ndarray or Torch tensor), always
+        floating-point with size ``(N, M)``:
+        - if ``boxes1`` has a floating-point dtype, the same dtype is used.
+        - if ``boxes1`` has an integer dtype, the result is returned as ``torch.float32``.
 
     """
 
@@ -842,16 +845,18 @@ def box_iou(boxes1: NdarrayOrTensor, boxes2: NdarrayOrTensor) -> NdarrayOrTensor
 
     inter, union = _box_inter_union(boxes1_t, boxes2_t, compute_dtype=COMPUTE_DTYPE)
 
-    # compute IoU and convert back to original box_dtype
+    # compute IoU and convert back to original box_dtype or torch.float32
     iou_t = inter / (union + torch.finfo(COMPUTE_DTYPE).eps)  # (N,M)
+    if not box_dtype.is_floating_point:
+        box_dtype = COMPUTE_DTYPE
     iou_t = iou_t.to(dtype=box_dtype)
 
     # check if NaN or Inf
     if torch.isnan(iou_t).any() or torch.isinf(iou_t).any():
         raise ValueError("Box IoU is NaN or Inf.")
 
     # convert tensor back to numpy if needed
-    iou, *_ = convert_to_dst_type(src=iou_t, dst=boxes1)
+    iou, *_ = convert_to_dst_type(src=iou_t, dst=boxes1, dtype=box_dtype)
     return iou
 
 
@@ -867,7 +872,10 @@ def box_giou(boxes1: NdarrayOrTensor, boxes2: NdarrayOrTensor) -> NdarrayOrTenso
         boxes2: bounding boxes, Mx4 or Mx6 torch tensor or ndarray. The box mode is assumed to be ``StandardMode``
 
     Returns:
-        GIoU, with size of (N,M) and same data type as ``boxes1``
+        An array/tensor matching the container type of ``boxes1`` (NumPy ndarray or Torch tensor), always
+        floating-point with size ``(N, M)``:
+        - if ``boxes1`` has a floating-point dtype, the same dtype is used.
+        - if ``boxes1`` has an integer dtype, the result is returned as ``torch.float32``.
 
     Reference:
         https://giou.stanford.edu/GIoU.pdf
@@ -904,12 +912,15 @@ def box_giou(boxes1: NdarrayOrTensor, boxes2: NdarrayOrTensor) -> NdarrayOrTenso
 
     # GIoU
     giou_t = iou - (enclosure - union) / (enclosure + torch.finfo(COMPUTE_DTYPE).eps)
+    if not box_dtype.is_floating_point:
+        box_dtype = COMPUTE_DTYPE
     giou_t = giou_t.to(dtype=box_dtype)
+
     if torch.isnan(giou_t).any() or torch.isinf(giou_t).any():
         raise ValueError("Box GIoU is NaN or Inf.")
 
     # convert tensor back to numpy if needed
-    giou, *_ = convert_to_dst_type(src=giou_t, dst=boxes1)
+    giou, *_ = convert_to_dst_type(src=giou_t, dst=boxes1, dtype=box_dtype)
     return giou
 
 
@@ -925,7 +936,10 @@ def box_pair_giou(boxes1: NdarrayOrTensor, boxes2: NdarrayOrTensor) -> NdarrayOr
         boxes2: bounding boxes, same shape with boxes1. The box mode is assumed to be ``StandardMode``
 
     Returns:
-        paired GIoU, with size of (N,) and same data type as ``boxes1``
+        An array/tensor matching the container type of ``boxes1`` (NumPy ndarray or Torch tensor), always
+        floating-point with size ``(N, )``:
+        - if ``boxes1`` has a floating-point dtype, the same dtype is used.
+        - if ``boxes1`` has an integer dtype, the result is returned as ``torch.float32``.
 
     Reference:
         https://giou.stanford.edu/GIoU.pdf
@@ -982,12 +996,15 @@ def box_pair_giou(boxes1: NdarrayOrTensor, boxes2: NdarrayOrTensor) -> NdarrayOr
     enclosure = torch.prod(wh, dim=-1, keepdim=False)  # (N,)
 
     giou_t: torch.Tensor = iou - (enclosure - union) / (enclosure + torch.finfo(COMPUTE_DTYPE).eps)  # type: ignore
+    if not box_dtype.is_floating_point:
+        box_dtype = COMPUTE_DTYPE
     giou_t = giou_t.to(dtype=box_dtype)  # (N,spatial_dims)
+
     if torch.isnan(giou_t).any() or torch.isinf(giou_t).any():
         raise ValueError("Box GIoU is NaN or Inf.")
 
     # convert tensor back to numpy if needed
-    giou, *_ = convert_to_dst_type(src=giou_t, dst=boxes1)
+    giou, *_ = convert_to_dst_type(src=giou_t, dst=boxes1, dtype=box_dtype)
     return giou
 
 

monai/networks/blocks/patchembedding.py

@@ -12,21 +12,22 @@
 from __future__ import annotations
 
 from collections.abc import Sequence
+from typing import Optional
 
 import numpy as np
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 from torch.nn import LayerNorm
 
-from monai.networks.blocks.pos_embed_utils import build_sincos_position_embedding
+from monai.networks.blocks.pos_embed_utils import build_fourier_position_embedding, build_sincos_position_embedding
 from monai.networks.layers import Conv, trunc_normal_
 from monai.utils import ensure_tuple_rep, optional_import
 from monai.utils.module import look_up_option
 
 Rearrange, _ = optional_import("einops.layers.torch", name="Rearrange")
 SUPPORTED_PATCH_EMBEDDING_TYPES = {"conv", "perceptron"}
-SUPPORTED_POS_EMBEDDING_TYPES = {"none", "learnable", "sincos"}
+SUPPORTED_POS_EMBEDDING_TYPES = {"none", "learnable", "sincos", "fourier"}
 
 
 class PatchEmbeddingBlock(nn.Module):
@@ -53,6 +54,7 @@ def __init__(
         pos_embed_type: str = "learnable",
         dropout_rate: float = 0.0,
         spatial_dims: int = 3,
+        pos_embed_kwargs: Optional[dict] = None,
     ) -> None:
         """
         Args:
@@ -65,6 +67,8 @@ def __init__(
             pos_embed_type: position embedding layer type.
             dropout_rate: fraction of the input units to drop.
             spatial_dims: number of spatial dimensions.
+            pos_embed_kwargs: additional arguments for position embedding. For `sincos`, it can contain
+                              `temperature` and for fourier it can contain `scales`.
         """
 
         super().__init__()
@@ -105,6 +109,8 @@ def __init__(
         self.position_embeddings = nn.Parameter(torch.zeros(1, self.n_patches, hidden_size))
         self.dropout = nn.Dropout(dropout_rate)
 
+        pos_embed_kwargs = {} if pos_embed_kwargs is None else pos_embed_kwargs
+
         if self.pos_embed_type == "none":
             pass
         elif self.pos_embed_type == "learnable":
@@ -114,7 +120,17 @@ def __init__(
             for in_size, pa_size in zip(img_size, patch_size):
                 grid_size.append(in_size // pa_size)
 
-            self.position_embeddings = build_sincos_position_embedding(grid_size, hidden_size, spatial_dims)
+            self.position_embeddings = build_sincos_position_embedding(
+                grid_size, hidden_size, spatial_dims, **pos_embed_kwargs
+            )
+        elif self.pos_embed_type == "fourier":
+            grid_size = []
+            for in_size, pa_size in zip(img_size, patch_size):
+                grid_size.append(in_size // pa_size)
+
+            self.position_embeddings = build_fourier_position_embedding(
+                grid_size, hidden_size, spatial_dims, **pos_embed_kwargs
+            )
         else:
             raise ValueError(f"pos_embed_type {self.pos_embed_type} not supported.")
 

monai/networks/blocks/pos_embed_utils.py

@@ -18,7 +18,7 @@
 import torch
 import torch.nn as nn
 
-__all__ = ["build_sincos_position_embedding"]
+__all__ = ["build_fourier_position_embedding", "build_sincos_position_embedding"]
 
 
 # From PyTorch internals
@@ -32,6 +32,59 @@ def parse(x):
     return parse
 
 
+def build_fourier_position_embedding(
+    grid_size: Union[int, List[int]], embed_dim: int, spatial_dims: int = 3, scales: Union[float, List[float]] = 1.0
+) -> torch.nn.Parameter:
+    """
+    Builds a (Anistropic) Fourier feature position embedding based on the given grid size, embed dimension,
+    spatial dimensions, and scales. The scales control the variance of the Fourier features, higher values make distant
+    points more distinguishable.
+    Position embedding is made anistropic by allowing setting different scales for each spatial dimension.
+        Reference: https://arxiv.org/abs/2509.02488
+
+    Args:
+        grid_size (int | List[int]): The size of the grid in each spatial dimension.
+        embed_dim (int): The dimension of the embedding.
+        spatial_dims (int): The number of spatial dimensions (2 for 2D, 3 for 3D).
+        scales (float | List[float]): The scale for every spatial dimension. If a single float is provided,
+                              the same scale is used for all dimensions.
+
+    Returns:
+        pos_embed (nn.Parameter): The Fourier feature position embedding as a fixed parameter.
+    """
+
+    to_tuple = _ntuple(spatial_dims)
+    grid_size_t = to_tuple(grid_size)
+    if len(grid_size_t) != spatial_dims:
+        raise ValueError(f"Length of grid_size ({len(grid_size_t)}) must be the same as spatial_dims.")
+
+    if embed_dim % 2 != 0:
+        raise ValueError("embed_dim must be even for Fourier position embedding")
+
+    # Ensure scales is a tensor of shape (spatial_dims,)
+    if isinstance(scales, float):
+        scales_tensor = torch.full((spatial_dims,), scales, dtype=torch.float)
+    elif isinstance(scales, (list, tuple)):
+        if len(scales) != spatial_dims:
+            raise ValueError(f"Length of scales {len(scales)} does not match spatial_dims {spatial_dims}")
+        scales_tensor = torch.tensor(scales, dtype=torch.float)
+    else:
+        raise TypeError(f"scales must be float or list of floats, got {type(scales)}")
+
+    gaussians = torch.randn(embed_dim // 2, spatial_dims, dtype=torch.float32) * scales_tensor
+
+    position_indices = [torch.linspace(0, 1, x, dtype=torch.float32) for x in grid_size_t]
+    positions = torch.stack(torch.meshgrid(*position_indices, indexing="ij"), dim=-1)
+    positions = positions.flatten(end_dim=-2)
+
+    x_proj = (2.0 * torch.pi * positions) @ gaussians.T
+
+    pos_emb = torch.cat([torch.sin(x_proj), torch.cos(x_proj)], dim=-1)
+    pos_emb = nn.Parameter(pos_emb[None, :, :], requires_grad=False)
+
+    return pos_emb
+
+
 def build_sincos_position_embedding(
     grid_size: Union[int, List[int]], embed_dim: int, spatial_dims: int = 3, temperature: float = 10000.0
 ) -> torch.nn.Parameter:

monai/transforms/__init__.py

@@ -506,7 +506,7 @@
     ZoomDict,
 )
 from .spatial.functional import spatial_resample
-from .traits import LazyTrait, MultiSampleTrait, RandomizableTrait, ThreadUnsafe
+from .traits import LazyTrait, MultiSampleTrait, RandomizableTrait, ReduceTrait, ThreadUnsafe
 from .transform import LazyTransform, MapTransform, Randomizable, RandomizableTransform, Transform, apply_transform
 from .utility.array import (
     AddCoordinateChannels,
@@ -521,6 +521,7 @@
     EnsureChannelFirst,
     EnsureType,
     FgBgToIndices,
+    FlattenSequence,
     Identity,
     ImageFilter,
     IntensityStats,
@@ -593,6 +594,9 @@
     FgBgToIndicesd,
     FgBgToIndicesD,
     FgBgToIndicesDict,
+    FlattenSequenced,
+    FlattenSequenceD,
+    FlattenSequenceDict,
     FlattenSubKeysd,
     FlattenSubKeysD,
     FlattenSubKeysDict,