fix shapes for batched outputs

examples/extract_from_single_2d.py

@@ -0,0 +1,17 @@
+import numpy as np
+import torch
+
+from torch_subpixel_crop import subpixel_crop_2d
+from skimage import data
+
+image = torch.tensor(data.binary_blobs(length=512, n_dim=2)).float()
+positions = torch.tensor(np.random.uniform(low=0, high=511, size=(100, 2))).float()
+
+crops = subpixel_crop_2d(
+    image=image,
+    positions=positions,
+    sidelength=32
+)
+
+# (100, 32, 32)
+print(crops.shape)

src/torch_subpixel_crop/subpixel_crop_2d.py

@@ -9,7 +9,7 @@
 
 
 def subpixel_crop_2d(
-    images: torch.Tensor, positions: torch.Tensor, sidelength: int,
+    image: torch.Tensor, positions: torch.Tensor, sidelength: int,
 ):
     """Extract square patches from 2D images at positions with subpixel precision.
 
@@ -18,7 +18,7 @@ def subpixel_crop_2d(
 
     Parameters
     ----------
-    images: torch.Tensor
+    image: torch.Tensor
         `(b, h, w)` or `(h, w)` array of 2D images.
     positions: torch.Tensor
         `(..., b, 2)` or `(..., 2)` array of coordinates for patch centers.
@@ -28,12 +28,18 @@ def subpixel_crop_2d(
     Returns
     -------
     patches: torch.Tensor
-        `(..., b, sidelength, sidelength)` array of patches from `images` with their
-        centers (DC components of fftshifted DFTs at `positions`.
+        `(..., b, sidelength, sidelength)` or `(..., sidelength, sidelength)` array
+         of patches from `images` with their centers at `positions`.
     """
-    if images.ndim == 2:
-        images = einops.rearrange(images, 'h w -> 1 h w')
+    # handling batched input
+    if image.ndim == 2:
+        input_images_are_batched = False
+        image = einops.rearrange(image, 'h w -> 1 h w')
         positions = einops.rearrange(positions, '... yx -> ... 1 yx')
+    else:
+        input_images_are_batched = True
+
+    # setup coordinates and extract
     positions, ps = einops.pack([positions], pattern='* t yx')
     positions = einops.rearrange(positions, 'b t yx -> t b yx')
     patches = einops.rearrange(
@@ -44,11 +50,15 @@ def subpixel_crop_2d(
                 output_image_sidelength=sidelength
             )
             for _image, _positions
-            in zip(images, positions)
+            in zip(image, positions)
         ],
         pattern='t b h w -> b t h w'
     )
     [patches] = einops.unpack(patches, pattern='* t h w', packed_shapes=ps)
+
+    # unbatch output if input images weren't batched
+    if input_images_are_batched is False:
+        patches = einops.rearrange(patches, pattern='... 1 h w -> ... h w')
     return patches
 
 
@@ -86,5 +96,5 @@ def _extract_patches_from_single_image(
     patches = einops.rearrange(patches, 'b 1 h w -> b h w')
 
     # phase shift to center images
-    patches = fourier_shift_image_2d(images=patches, shifts=shifts)
+    patches = fourier_shift_image_2d(image=patches, shifts=shifts)
     return patches