feat: add int8, int16 to downsample_with_averaging

.github/workflows/build_wheel.yml

@@ -28,12 +28,16 @@ jobs:
         uses: docker/setup-qemu-action@v1
 
       - name: Build wheels
-        uses: pypa/cibuildwheel@v2.19.1
+        uses: joerick/cibuildwheel@v2.22.0
         # to supply options, put them in 'env', like:
         env:
-          CIBW_ARCHS_LINUX: ${{matrix.arch}}
-          CIBW_BEFORE_BUILD: pip install numpy setuptools wheel cython
+          CIBW_BEFORE_BUILD: pip install numpy pybind11 setuptools wheel pkginfo twine
+          CIBW_ARCHS_MACOS: "x86_64 arm64"
+          CIBW_ARCHS: auto64
 
-      - uses: actions/upload-artifact@v2
+      - name: Upload built wheels
+        uses: actions/upload-artifact@v4
         with:
+          name: built-wheels-${{ matrix.os }}-${{ matrix.arch }}
           path: ./wheelhouse/*.whl
+          if-no-files-found: warn

.github/workflows/run_tests.yml

@@ -14,7 +14,7 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-20.04, macos-latest, windows-2019]
+        os: [ubuntu-latest, macos-latest, windows-latest]
         python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
 
     steps:

README.md

@@ -14,7 +14,12 @@ img = load_3d_em_stack()
 img_pyramid = tinybrain.downsample_with_averaging(img, factor=(2,2,1), num_mips=5, sparse=False)
 
 labels = load_3d_labels()
-label_pyramid = tinybrain.downsample_segmentation(labels, factor=(2,2,1), num_mips=5, sparse=False))
+label_pyramid = tinybrain.downsample_segmentation(labels, factor=(2,2,1), num_mips=5, sparse=False)
+
+# We also have a few other types
+img_pyramid = tinybrain.downsample_with_min_pooling(image, factor=(2,2,1), num_mips=5)
+img_pyramid = tinybrain.downsample_with_max_pooling(image, factor=(2,2,1), num_mips=5)
+img_pyramid = tinybrain.downsample_with_striding(image, factor=(2,2,1), num_mips=5)
 ```
 
 ## Installation 
@@ -27,7 +32,7 @@ pip install tinybrain
 ## Motivation
 
 Image hierarchy generation in connectomics uses a few different techniques for
-visualizing data, but predominantly we create image pyramids of uint8 grayscale images using 2x2 average pooling and of uint8 to uint64 segmentation labels using 2x2 mode pooling. When images become very large and people wish to visualze upper mip levels using three axes at once, it becomes desirable to perform 2x2x2 downsamples to maintain isotropy.
+visualizing data, but predominantly we create image pyramids of uint8 grayscale images using 2x2 average pooling and of uint8 to uint64 segmentation labels using 2x2 mode pooling. When images become very large and people wish to visualize upper mip levels using three axes at once, it becomes desirable to perform 2x2x2 downsamples to maintain isotropy.
 
 It's possible to compute both of these using numpy, however as multiple packages found it useful to copy the downsample functions, it makes sense to formalize these functions into a seperate library located on PyPI.
 

automated_test.py

@@ -125,7 +125,7 @@ def test_even_odd2d():
   assert np.array_equal(oddimg, ans3x3x3)
   assert np.array_equal(oddimgf, ans3x3x3)
 
-@pytest.mark.parametrize("dtype", (np.uint8, np.uint16, np.float32, np.float64))
+@pytest.mark.parametrize("dtype", (np.int8, np.int16, np.uint8, np.uint16, np.float32, np.float64))
 @pytest.mark.parametrize("sparse", [False, True])
 def test_accelerated_vs_numpy_avg_pooling_2x2x1(dtype, sparse):
   image = np.random.randint(0,255, size=(512, 512, 6), dtype=np.uint8).astype(dtype, copy=False)
@@ -147,7 +147,7 @@ def test_accelerated_vs_numpy_avg_pooling_2x2x1(dtype, sparse):
 
   assert np.all(mips[-1] == npimg)
 
-@pytest.mark.parametrize("dtype", (np.uint8, np.uint16, np.float32, np.float64))
+@pytest.mark.parametrize("dtype", (np.int8, np.int16, np.uint8, np.uint16, np.float32, np.float64))
 def test_accelerated_vs_numpy_avg_pooling_2x2x1_simple_sparse(dtype):
   for x in [0,1]:
     for y in [0,1]:
@@ -189,7 +189,7 @@ def test_accelerated_vs_numpy_avg_pooling_2x2x1_simple_sparse(dtype):
   assert np.all(res[1] == ans)
 
 
-@pytest.mark.parametrize("dtype", (np.uint8, np.uint16, np.float32, np.float64))
+@pytest.mark.parametrize("dtype", (np.int8, np.int16, np.uint8, np.uint16, np.float32, np.float64))
 @pytest.mark.parametrize("sx", (6,7,1024,1025))
 @pytest.mark.parametrize("sy", (6,7,1024,1025))
 @pytest.mark.parametrize("sz", (4,5,32,33))

tinybrain/accelerated.hpp

@@ -1,5 +1,5 @@
 /*
-Copyright (C) 2019, William Silversmith
+Copyright (C) 2019,2025 William Silversmith
 
 This program is free software: you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -359,7 +359,7 @@ T* _average_pooling_2x2_single_mip(
                   + static_cast<T2>(d)
                 ) / static_cast<T2>(
                     std::max(
-                      static_cast<T>((a > 0) + (b > 0) + (c > 0) + (d > 0)),
+                      static_cast<T>((a != 0) + (b != 0) + (c != 0) + (d != 0)),
                       static_cast<T>(1)
                     )
                   )
@@ -375,7 +375,7 @@ T* _average_pooling_2x2_single_mip(
                   + static_cast<T2>(b)
                 ) / static_cast<T2>(
                   std::max(
-                    static_cast<T>((a > 0) + (b > 0)),
+                    static_cast<T>((a != 0) + (b != 0)),
                     static_cast<T>(1)
                 )
               ));
@@ -931,8 +931,17 @@ U* denominator_2x2x2(
 
 template <typename T, typename U>
 inline void render_image(T* accum, U* oimg, const uint32_t bitshift, const size_t ovoxels) {
-  for (size_t i = 0; i < ovoxels; i++) {
-    oimg[i] = static_cast<U>(accum[i] >> bitshift);
+  if constexpr (std::is_signed<T>::value) {
+    for (size_t i = 0; i < ovoxels; i++) {
+      oimg[i] = (accum[i] < 0) 
+        ? -static_cast<U>(std::abs(accum[i]) >> bitshift)
+        : static_cast<U>(accum[i] >> bitshift);
+    }
+  } 
+  else {
+    for (size_t i = 0; i < ovoxels; i++) {
+      oimg[i] = static_cast<U>(accum[i] >> bitshift);
+    }
   }
 }