monai-weekly 1.7.dev2644__tar.gz → 1.7.dev2645__tar.gz

{monai_weekly-1.7.dev2644 → monai_weekly-1.7.dev2645}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: monai-weekly
-Version: 1.7.dev2644
+Version: 1.7.dev2645
 Summary: AI Toolkit for Healthcare Imaging
 Author-email: MONAI Consortium <monai.contact@gmail.com>
 License: Apache License 2.0

{monai_weekly-1.7.dev2644 → monai_weekly-1.7.dev2645}/monai/__init__.py

@@ -137,4 +137,4 @@ except Exception:
 
     if MONAIEnvVars.debug():
         raise
-__commit_id__ = "827fa24fef4f474f70ead141412a2a0675990684"
+__commit_id__ = "c5bddb21ccaaf3654a87c65ede6d2c7bd82e4128"

{monai_weekly-1.7.dev2644 → monai_weekly-1.7.dev2645}/monai/_version.py

@@ -8,11 +8,11 @@ import json
 
 version_json = '''
 {
- "date": "2026-11-01T02:09:43+0000",
+ "date": "2026-11-08T02:09:15+0000",
  "dirty": false,
  "error": null,
- "full-revisionid": "47589b4b36aee36f5956dc5d440c445c3a7b1e2a",
- "version": "1.7.dev2644"
+ "full-revisionid": "ed1d0c3cb86b2cfe223a7c845bec698dedbfa614",
+ "version": "1.7.dev2645"
 }
 '''  # END VERSION_JSON
 

{monai_weekly-1.7.dev2644 → monai_weekly-1.7.dev2645}/monai/networks/nets/quicknat.py

@@ -128,15 +128,15 @@ class ConvConcatDenseBlock(ConvDenseBlock):
     Every convolutional layer is preceded by a batch-normalization layer and a Rectifier Linear Unit (ReLU) layer.
     The first two convolutional layers are followed by a concatenation layer that concatenates
     the input feature map with outputs of the current and previous convolutional blocks.
-    Kernel size of two convolutional layers kept small to limit number of paramters.
+    Kernel size of two convolutional layers kept small to limit number of parameters.
     Appropriate padding is provided so that the size of feature maps before and after convolution remains constant.
     The output channels for each convolution layer is set to 64, which acts as a bottle- neck for feature map selectivity.
     The input channel size is variable, depending on the number of dense connections.
     The third convolutional layer is also preceded by a batch normalization and ReLU,
     but has a 1 * 1 kernel size to compress the feature map size to 64.
     Args:
-        in_channles: variable depending on depth of the network
-        seLayer: Squeeze and Excite block to be included, defaults to None, valid options are {'NONE', 'CSE', 'SSE', 'CSSE'},
+        in_channels: variable depending on depth of the network
+        se_layer: Squeeze and Excite block to be included, defaults to None, valid options are {'NONE', 'CSE', 'SSE', 'CSSE'},
         dropout_layer: Dropout block to be included, defaults to None.
     :return: forward passed tensor
     """
@@ -343,11 +343,11 @@ class Quicknat(nn.Module):
         num_filters: number of output channels for each convolutional layer in a Dense Block.
         kernel_size: size of the kernel of each convolutional layer in a Dense Block.
         kernel_c: convolution kernel size of classifier block kernel.
-        stride_convolution: convolution stride. Defaults to 1.
+        stride_conv: convolution stride. Defaults to 1.
         pool: kernel size of the pooling layer,
         stride_pool: stride for the pooling layer.
         se_block: Squeeze and Excite block type to include. Use ``None`` or ``"None"`` to disable it; valid enabled values are ``"CSE"``, ``"SSE"``, and ``"CSSE"``.
-        droup_out: dropout ratio. Defaults to no dropout.
+        drop_out: dropout ratio. Defaults to no dropout.
         act: activation type and arguments. Defaults to PReLU.
         norm: feature normalization type and arguments. Defaults to instance norm.
         adn_ordering: a string representing the ordering of activation (A), normalization (N), and dropout (D).

{monai_weekly-1.7.dev2644 → monai_weekly-1.7.dev2645}/monai/transforms/spatial/functional.py

@@ -16,6 +16,7 @@ from __future__ import annotations
 
 import math
 import warnings
+from collections.abc import Sequence
 from enum import Enum
 
 import numpy as np
@@ -106,6 +107,41 @@ def _maybe_new_metatensor(img, dtype=None, device=None):
     )
 
 
+def _cancel_align_corners_scaling(
+    xform: torch.Tensor, in_spatial_size: Sequence[int], out_spatial_size: Sequence[int]
+) -> torch.Tensor:
+    """
+    Pre-compensate a voxel-space affine for the scalings applied again downstream when
+    ``align_corners=True`` and the compiled ``Resample`` path is used.
+
+    ``xform`` is already expressed in voxel space. However, ``Affine`` passes it through ``AffineGrid``,
+    which applies a ``d_out / (d_out - 1)`` scaling per spatial dimension, and the compiled ``Resample``
+    applies a ``(d_in - 1) / d_in`` scaling before ``grid_pull``. This function returns
+    ``T @ S_inv @ xform @ D_inv``, where ``S_inv`` and ``D_inv`` cancel those two scalings and ``T`` is a
+    translation by -0.5 voxel that re-centres the result, so that ``grid_pull`` receives the original
+    ``xform`` unchanged.
+
+    Dimensions are clamped to a minimum of 2 to avoid division by zero for singleton axes.
+
+    Args:
+        xform: homogeneous affine of shape ``(rank + 1, rank + 1)`` in voxel space.
+        in_spatial_size: spatial shape of the input image.
+        out_spatial_size: spatial shape of the resampled output.
+
+    Returns:
+        the compensated homogeneous affine, same shape, dtype and device as ``xform``.
+    """
+    rank = len(xform) - 1
+    eye = torch.eye(rank + 1, dtype=xform.dtype, device=xform.device)
+    s_inv, d_inv, translate = eye.clone(), eye.clone(), eye.clone()
+    for i in range(rank):
+        d_in, d_out = max(2, int(in_spatial_size[i])), max(2, int(out_spatial_size[i]))
+        s_inv[i, i] = d_in / (d_in - 1)
+        d_inv[i, i] = (d_out - 1) / d_out
+        translate[i, -1] = -0.5
+    return translate @ s_inv @ xform @ d_inv
+
+
 def spatial_resample(
     img, dst_affine, spatial_size, mode, padding_mode, align_corners, dtype_pt, lazy, transform_info
 ) -> torch.Tensor:
@@ -216,6 +252,8 @@ def spatial_resample(
     if isinstance(mode, int) or _use_compiled:
         dst_xform = create_translate(spatial_rank, [float(d - 1) / 2 for d in spatial_size])
         xform = xform @ convert_to_dst_type(dst_xform, xform)[0]
+        if align_corners and not isinstance(mode, int):
+            xform = _cancel_align_corners_scaling(xform, in_sp_size, spatial_size)
         affine_xform = monai.transforms.Affine(
             affine=xform,
             spatial_size=spatial_size,

{monai_weekly-1.7.dev2644 → monai_weekly-1.7.dev2645}/monai_weekly.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: monai-weekly
-Version: 1.7.dev2644
+Version: 1.7.dev2645
 Summary: AI Toolkit for Healthcare Imaging
 Author-email: MONAI Consortium <monai.contact@gmail.com>
 License: Apache License 2.0