add Solarize2D

CHANEGLOG.md

@@ -37,6 +37,17 @@
       return SimpleRNNState(jnp.zeros([cell.hidden_features]))
   ```
 
+- `MultiHeadAttention`
+- `BatchNorm`
+- `RandomHorizontalShear2D`
+- `RandomPerspective2D`
+- `RandomRotate2D`
+- `RandomVerticalShear2D`
+- `Rotate2D`
+- `VerticalShear2D`
+- `Pixelate2D`
+- `Solarize2D`
+
 ### Deprecations
 
 - `Bilinear` is deprecated, use `Multilinear((in1_features, in2_features), out_features)`

serket/nn/__init__.py

@@ -102,6 +102,7 @@
     RandomRotate2D,
     RandomVerticalShear2D,
     Rotate2D,
+    Solarize2D,
     VerticalShear2D,
 )
 from .linear import FNN, MLP, Embedding, GeneralLinear, Identity, Linear, Multilinear
@@ -279,6 +280,7 @@
     "RandomRotate2D",
     "RandomVerticalShear2D",
     "Rotate2D",
+    "Solarize2D",
     "VerticalShear2D",
     # pooling
     "AdaptiveAvgPool1D",

serket/nn/image.py

@@ -908,6 +908,53 @@ def spatial_ndim(self) -> int:
         return 2
 
 
+def solarize(
+    image: jax.Array,
+    threshold: float | int,
+    max_val: float | int,
+) -> jax.Array:
+    """Inverts all values above a given threshold."""
+    return jnp.where(image < threshold, image, max_val - image)
+
+
+@sk.autoinit
+class Solarize2D(sk.TreeClass):
+    """Inverts all values above a given threshold.
+
+    Args:
+        threshold: The threshold value above which to invert.
+        max_val: The maximum value of the image. e.g. 255 for uint8 images.
+            1.0 for float images. default: 1.0
+
+    Example:
+        >>> import serket as sk
+        >>> import jax.numpy as jnp
+        >>> x = jnp.arange(1, 26).reshape(1, 5, 5)
+        >>> layer = sk.nn.Solarize2D(threshold=10, max_val=25)
+        >>> print(layer(x))
+        [[[ 1  2  3  4  5]
+          [ 6  7  8  9 15]
+          [14 13 12 11 10]
+          [ 9  8  7  6  5]
+          [ 4  3  2  1  0]]]
+
+    Reference:
+        - https://github.com/tensorflow/models/blob/v2.13.1/official/vision/ops/augment.py#L804-L809
+    """
+
+    threshold: float
+    max_val: float = 1.0
+
+    @ft.partial(validate_spatial_ndim, attribute_name="spatial_ndim")
+    def __call__(self, x: jax.Array, **k) -> jax.Array:
+        threshold, max_val = jax.lax.stop_gradient((self.threshold, self.max_val))
+        return solarize(x, threshold, max_val)
+
+    @property
+    def spatial_ndim(self) -> int:
+        return 2
+
+
 @tree_eval.def_eval(RandomContrast2D)
 @tree_eval.def_eval(RandomRotate2D)
 @tree_eval.def_eval(RandomHorizontalShear2D)

tests/test_image_filter.py

@@ -18,7 +18,7 @@
 import numpy.testing as npt
 import pytest
 
-from serket.nn import AvgBlur2D, FFTFilter2D, Filter2D, GaussianBlur2D
+from serket.nn import AvgBlur2D, FFTFilter2D, Filter2D, GaussianBlur2D, Solarize2D
 
 
 def test_AvgBlur2D():
@@ -99,3 +99,22 @@ def test_filter2d():
     layer2 = FFTFilter2D(in_features=1, kernel=jnp.ones([3, 3]) / 9.0)
 
     npt.assert_allclose(layer(x), layer2(x), atol=1e-4)
+
+
+def test_solarize2d():
+    x = jnp.arange(1, 26).reshape(1, 5, 5)
+    layer = Solarize2D(threshold=10, max_val=25)
+    npt.assert_allclose(
+        layer(x),
+        jnp.array(
+            [
+                [
+                    [1, 2, 3, 4, 5],
+                    [6, 7, 8, 9, 15],
+                    [14, 13, 12, 11, 10],
+                    [9, 8, 7, 6, 5],
+                    [4, 3, 2, 1, 0],
+                ]
+            ]
+        ),
+    )