Add changelogs

.github/workflows/test_pull_requests.yml

@@ -21,7 +21,7 @@ jobs:
         os: [ubuntu-latest, macos-latest, windows-latest]
         # see supported versions at
         # https://raw.githubusercontent.com/actions/python-versions/master/versions-manifest.json
-        python-version: ['3.7', '3.8', '3.9', '3.10', '3.11', '3.12']
+        python-version: ['3.8', '3.9', '3.10', '3.11', '3.12']
         # test only the latest 3.x on mac
         # test only the latest 3.x on windows
         exclude:
@@ -85,27 +85,27 @@ jobs:
         echo "::set-output name=date::$(/bin/date -u "+%Y%m%d")"
       shell: bash
 
-    - uses: actions/cache@v1
+    - uses: actions/cache@v4
       if: startsWith(runner.os, 'Linux')
       with:
         path: ~/.cache/pip
-        key: ${{ runner.os }}-pip-${{ steps.get-date.outputs.date }}-${{ hashFiles('**/requirements.txt') }}
+        key: ${{ runner.os }}-pip-${{ steps.get-date.outputs.date }}-${{ hashFiles('setup.py') }}
         restore-keys: |
           ${{ runner.os }}-pip-
 
-    - uses: actions/cache@v1
+    - uses: actions/cache@v4
       if: startsWith(runner.os, 'macOS')
       with:
         path: ~/Library/Caches/pip
-        key: ${{ runner.os }}-pip-${{ steps.get-date.outputs.date }}-${{ hashFiles('**/requirements.txt') }}
+        key: ${{ runner.os }}-pip-${{ steps.get-date.outputs.date }}-${{ hashFiles('setup.py') }}
         restore-keys: |
           ${{ runner.os }}-pip-
 
-    - uses: actions/cache@v1
+    - uses: actions/cache@v4
       if: startsWith(runner.os, 'Windows')
       with:
         path: ~\AppData\Local\pip\Cache
-        key: ${{ runner.os }}-pip-${{ steps.get-date.outputs.date }}-${{ hashFiles('**/requirements.txt') }}
+        key: ${{ runner.os }}-pip-${{ steps.get-date.outputs.date }}-${{ hashFiles('setup.py') }}
         restore-keys: |
           ${{ runner.os }}-pip-
 

changelogs/master/changed/20241202_change_scikit-image_support.md

@@ -0,0 +1,5 @@
+# Requires scikit-image >= 0.18.0
+
+Changes in the tests require a different minimal
+`scikit-image` version. Older versions might work
+but will be untested.

changelogs/master/changed/20241216_change_supported_python.md

@@ -0,0 +1 @@
+# Support Python >=3.7,<=3.12

changelogs/master/changed/20241217_add_opencv4_support.md

@@ -0,0 +1 @@
+# Add OpenCV4 suport

changelogs/master/fixed/20240918_fix_numpy_2_sctypes.md

@@ -0,0 +1,6 @@
+# Fix broken dtypes for numpy 2.0 #856
+
+`np.sctypes` was removed in the NumPy 2.0.
+All occurences were replaced with a set of 
+the specific NumPy types.
+

changelogs/master/fixed/20241025_fix_cval_is_depreciated.md

@@ -0,0 +1,5 @@
+# Replace default (and depreciated) cval with replacement
+
+In `skimage.measure.block_reduce` in `imgaug.pool` the constant value is
+set by depreciated `cval` variable althought replacement `pad_cval` is
+available.

changelogs/master/fixed/20241111_fix_failing_test_seed.md

@@ -0,0 +1,8 @@
+# Relax test condition for unlucky seed choice 
+
+After trying back and forth I came to the conclusion that the 
+`test/augmenters/test_geometric.py:_test_image_cbaoi_alignment`fails
+sindce target value is highly dependend on the seed value. With `seed=1` for `iaa.ElasticTransformation` `count_bad` goes to 0.
+
+This provides the conclusion even with a changed seed it cannot be granted that the test will pass for future random generators. Thus
+a relaxed target value might be the better choice.

changelogs/master/fixed/20241127_fix_zero_sized_arrays.md

@@ -0,0 +1,7 @@
+# OpenCV 3 does not support zero-sized arrays
+
+NumPy 1.23 can create zero-sized arrays which
+cannot be handled by OpenCV in `_invert_uint8_subtract_`
+in `imgaug/augmenters/arithmetic.py`. If a zero-sized
+array is detected directly return it without further 
+processing.

changelogs/master/fixed/20241204_fix_improved_skimage_cvals.md

@@ -0,0 +1,5 @@
+# Changed behavior of `_augment_images_by_samples`
+
+The cval test became obsolete for skimage > 0.19 since the piecewise
+tested affine transform does not produce empty pixels anymore. The
+test for cval == 0 only passed due to uint overflows.

changelogs/master/fixed/20241204_fix_scikit-image_warp.md

@@ -0,0 +1,7 @@
+# Mitigate different behavior for scikit-image's transform.warp
+
+Prior to `scikit-image=0.19.0` `transform.warp` converted float16 images
+implicitly to float64 prior to the calculation. This has changed such
+that we need to convert the image to a supported floating point format
+ourselves in `_warp_affine_arr_skimage` and `_augment_images_by_samples` 
+in `imgaug/augmenters/geometric.py`.

changelogs/master/fixed/20241206_fix_skimage_determinism.md

@@ -0,0 +1,15 @@
+# Create imagecorruptions fork
+
+Since `scikit-image >= 0.19.0` the determinism of some 
+functions cannot be controlled via manipulating the global
+NumPy random generator seed. This results in random behavior
+for tests involving `imagecorruptions` functions. 
+The only solutions is to add a seed parameter to the repective
+functions. A corresponding upstream pull request is pending but
+the changes are low such that a `imagecorruptions` fork is required.
+
+Another problem is the broken support of the `imagecorruptions`
+main branch for older `scikit-image` versions.
+
+In the future versions will merge the fork into the `imgaug` codebase and
+also require newer `scikit-image` versions.

changelogs/master/fixed/20241213_fix_add_numpy_126_support.md

@@ -0,0 +1,3 @@
+# Add NumPy 1.26 support
+
+Add the support of NumPy 1.26 to `test/test_random.py`.

changelogs/master/fixed/20241216_fix_failing_int128_tests.md

@@ -0,0 +1,9 @@
+# Delete different test routines for float128 systems 
+
+The test `test_unusual_channel_numbers` in 
+`test/augmenters/test_contrast.py` ran different test routines
+depending on the float128 support of the system.
+
+Yet, the main reason for the different behavior is the `params.draw_samples` call in `contrast.py:71` which converts a regular python int into a numpy array of int dtype and thus resulted in different floating point rounding errors.
+
+This behavior was not mirrored in the expected test output and thus resulted in different outcomes on systems with different floating point support.

changelogs/master/fixed/20241216_fix_seed_always_as_int.md

@@ -0,0 +1,3 @@
+# Make sure `random.seed` received int
+
+The multicore support was broken to to differend seed dtypes.

changelogs/master/improved/20241203_make_poly_test_consistent.md

@@ -0,0 +1,8 @@
+# Polygon test was inconsistent
+
+`test_alpha_is_080` in `test/augmentables/test_polys.py`
+hard-coded inconsistent behavior of `scikit-image` which
+was fixed with `scikit-image >=0.18.0`.
+
+Until this point, the polygon left and right boundaries were handled
+differently.

changelogs/master/refactored/20230413_replace_numpy_bool.md

@@ -0,0 +1,6 @@
+# Replace numpy bool with standard bool #832
+
+This commit replaces all occurences of the 
+dtype `np.bool` with `bool` from the python
+standard library. `np.bool` is marked as
+depreciated in `numpy >= 1.20`. 

changelogs/master/refactored/20230726_replace_numpy_complex.md

@@ -0,0 +1,4 @@
+Change numpy complex types #840
+
+The dtype `np.complex` is depreciated in numpy >= 1.20.
+All occurrences where replaced by `np.complex128`.

changelogs/master/refactored/20250306_github_actions_cache_ci.md

@@ -0,0 +1,9 @@
+# Changed github actions
+
+Version 1 of `actions/cache` is now depreciated. 
+The pipeline switched version 4.
+
+Since `requirements.txt` was deleted, the caching
+not relies on `setup.py` instead.
+
+Python 3.7 is not available on `ubuntu-latest` runners.

setup.py

@@ -100,7 +100,6 @@ def get_install_requirements(main_requires, alternative_requires):
         "Natural Language :: English",
         "Operating System :: OS Independent",
         "Programming Language :: Python :: 3",
-        "Programming Language :: Python :: 3.7",
         "Programming Language :: Python :: 3.8",
         "Programming Language :: Python :: 3.9",
         "Programming Language :: Python :: 3.10",