Fix dimension order in the image shape

impl/src/androidMain/kotlin/org/jetbrains/kotlinx/dl/impl/preprocessing/bitmap/Resize.kt

@@ -1,5 +1,5 @@
 /*
- * Copyright 2020-2022 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
+ * Copyright 2020-2023 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
  * Use of this source code is governed by the Apache 2.0 license that can be found in the LICENSE.txt file.
  */
 
@@ -35,8 +35,8 @@ public class Resize(
     public companion object {
         internal fun createOutputImageShape(inputShape: TensorShape, outputWidth: Int, outputHeight: Int): TensorShape {
             return when (inputShape.rank()) {
-                2 -> TensorShape(outputWidth.toLong(), outputHeight.toLong())
-                3 -> TensorShape(outputWidth.toLong(), outputHeight.toLong(), inputShape[2])
+                2 -> TensorShape(outputHeight.toLong(), outputWidth.toLong())
+                3 -> TensorShape(outputHeight.toLong(), outputWidth.toLong(), inputShape[2])
                 else -> throw IllegalArgumentException("Input shape is expected to be 2D or 3D")
             }
         }

impl/src/jvmMain/kotlin/org/jetbrains/kotlinx/dl/impl/preprocessing/image/Cropping.kt

@@ -1,5 +1,5 @@
 /*
- * Copyright 2020-2022 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
+ * Copyright 2020-2023 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
  * Use of this source code is governed by the Apache 2.0 license that can be found in the LICENSE.txt file.
  */
 
@@ -29,7 +29,7 @@ public class Cropping(
 ) : Operation<BufferedImage, BufferedImage> {
     override fun apply(input: BufferedImage): BufferedImage {
         val croppedImageShape = getOutputShape(input.getShape())
-        val (width, height, _) = croppedImageShape.dims()
+        val (height, width, _) = croppedImageShape.dims()
 
         return input.getSubimage(
             left, top,
@@ -39,12 +39,12 @@ public class Cropping(
     }
 
     override fun getOutputShape(inputShape: TensorShape): TensorShape {
-        val outputWidth = if (inputShape[0] == -1L) -1 else inputShape[0] - left - right
-        val outputHeight = if (inputShape[1] == -1L) -1 else inputShape[1] - top - bottom
+        val outputWidth = if (inputShape[1] == -1L) -1 else inputShape[1] - left - right
+        val outputHeight = if (inputShape[0] == -1L) -1 else inputShape[0] - top - bottom
 
         return when (inputShape.rank()) {
-            2 -> TensorShape(outputWidth, outputHeight)
-            3 -> TensorShape(outputWidth, outputHeight, inputShape[2])
+            2 -> TensorShape(outputHeight, outputWidth)
+            3 -> TensorShape(outputHeight, outputWidth, inputShape[2])
             else -> throw IllegalArgumentException("Cropping operation is applicable only to images with rank 2 or 3")
         }
     }

impl/src/jvmMain/kotlin/org/jetbrains/kotlinx/dl/impl/preprocessing/image/ImageUtil.kt

@@ -25,5 +25,5 @@ internal fun BufferedImage.copy(): BufferedImage {
 }
 
 internal fun BufferedImage.getShape(): TensorShape {
-    return TensorShape(width.toLong(), height.toLong(), colorModel.numComponents.toLong())
+    return TensorShape(height.toLong(), width.toLong(), colorModel.numComponents.toLong())
 }

impl/src/jvmMain/kotlin/org/jetbrains/kotlinx/dl/impl/preprocessing/image/Padding.kt

@@ -1,5 +1,5 @@
 /*
- * Copyright 2020-2022 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
+ * Copyright 2020-2023 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
  * Use of this source code is governed by the Apache 2.0 license that can be found in the LICENSE.txt file.
  */
 
@@ -47,12 +47,12 @@ public class Padding(
     }
 
     override fun getOutputShape(inputShape: TensorShape): TensorShape {
-        val outputWidth = if (inputShape[0] == -1L) -1 else inputShape[0] + left + right
-        val outputHeight = if (inputShape[1] == -1L) -1 else inputShape[1] + top + bottom
+        val outputWidth = if (inputShape[1] == -1L) -1 else inputShape[1] + left + right
+        val outputHeight = if (inputShape[0] == -1L) -1 else inputShape[0] + top + bottom
 
         return when (inputShape.rank()) {
-            2 -> TensorShape(outputWidth, outputHeight)
-            3 -> TensorShape(outputWidth, outputHeight, inputShape[2])
+            2 -> TensorShape(outputHeight, outputWidth)
+            3 -> TensorShape(outputHeight, outputWidth, inputShape[2])
             else -> throw IllegalArgumentException("Padding operation is supported only for 2D and 3D tensors")
         }
     }

impl/src/jvmMain/kotlin/org/jetbrains/kotlinx/dl/impl/preprocessing/image/Resize.kt

@@ -1,5 +1,5 @@
 /*
- * Copyright 2020-2022 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
+ * Copyright 2020-2023 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
  * Use of this source code is governed by the Apache 2.0 license that can be found in the LICENSE.txt file.
  */
 
@@ -65,8 +65,8 @@ public class Resize(
 
     override fun getOutputShape(inputShape: TensorShape): TensorShape {
         return when (inputShape.rank()) {
-            2 -> TensorShape(outputWidth.toLong(), outputHeight.toLong())
-            3 -> TensorShape(outputWidth.toLong(), outputHeight.toLong(), inputShape[2])
+            2 -> TensorShape(outputHeight.toLong(), outputWidth.toLong())
+            3 -> TensorShape(outputHeight.toLong(), outputWidth.toLong(), inputShape[2])
             else -> throw IllegalArgumentException("Resize operation is only supported for 2D and 3D tensors.")
         }
     }

impl/src/jvmTest/kotlin/org/jetbrains/kotlinx/dl/impl/preprocessing/PreprocessingFinalShapeTest.kt

@@ -1,5 +1,5 @@
 /*
- * Copyright 2020-2022 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
+ * Copyright 2020-2023 JetBrains s.r.o. and Kotlin Deep Learning project contributors. All Rights Reserved.
  * Use of this source code is governed by the Apache 2.0 license that can be found in the LICENSE.txt file.
  */
 
@@ -47,7 +47,7 @@ class PreprocessingFinalShapeTest {
                 bottom = 7
             }
             .toFloatArray { }
-        assertEquals(TensorShape(186, 188, 3), preprocess.getOutputShape(TensorShape(200, 200, 3)))
+        assertEquals(TensorShape(178, 186, 3), preprocess.getOutputShape(TensorShape(190, 200, 3)))
     }
 
     @Test
@@ -66,7 +66,7 @@ class PreprocessingFinalShapeTest {
                 bottom = 3
             }
             .toFloatArray { }
-        assertEquals(TensorShape(180, 180, 3), preprocess.getOutputShape(TensorShape(200, 200, 3)))
+        assertEquals(TensorShape(380, 180, 3), preprocess.getOutputShape(TensorShape(400, 200, 3)))
     }
 
     @Test
@@ -84,7 +84,7 @@ class PreprocessingFinalShapeTest {
                 bottom = 5
             }
             .toFloatArray { }
-        assertEquals(TensorShape(140, 90, 3), preprocess.getOutputShape(TensorShape(200, 200, 3)))
+        assertEquals(TensorShape(90, 140, 3), preprocess.getOutputShape(TensorShape(200, 200, 3)))
     }
 
     @Test
@@ -107,7 +107,7 @@ class PreprocessingFinalShapeTest {
                 right = 13
             }
             .toFloatArray { }
-        assertEquals(TensorShape(324, 212, 1), preprocess.getOutputShape(TensorShape(300, 200, 1)))
+        assertEquals(TensorShape(312, 224, 1), preprocess.getOutputShape(TensorShape(300, 200, 1)))
     }
 
     @Test
@@ -127,7 +127,7 @@ class PreprocessingFinalShapeTest {
         val image = BufferedImage(10, 20, BufferedImage.TYPE_3BYTE_BGR)
         val (_, actualShape) = preprocess.apply(image)
 
-        assertEquals(actualShape, preprocess.getOutputShape(TensorShape(10, 20, 1)))
+        assertEquals(actualShape, preprocess.getOutputShape(TensorShape(20, 10, 1)))
     }
 
     @Test
@@ -139,6 +139,6 @@ class PreprocessingFinalShapeTest {
         val image = BufferedImage(10, 20, BufferedImage.TYPE_3BYTE_BGR)
         val (_, actualShape) = preprocess.apply(image)
 
-        assertEquals(actualShape, preprocess.getOutputShape(TensorShape(10, 20, 3)))
+        assertEquals(actualShape, preprocess.getOutputShape(TensorShape(20, 10, 3)))
     }
 }

impl/src/jvmTest/kotlin/org/jetbrains/kotlinx/dl/impl/preprocessing/PreprocessingImageTest.kt

@@ -105,7 +105,7 @@ class PreprocessingImageTest {
         inputImage.setRGB(1, 1, Color.RED.rgb)
         val (imageFloats, tensorShape) = preprocess.apply(inputImage)
 
-        Assertions.assertEquals(TensorShape(9, 5, 3), tensorShape)
+        Assertions.assertEquals(TensorShape(5, 9, 3), tensorShape)
 
         val expectedImage = FloatArray(tensorShape.numElements().toInt()) { Color.GRAY.red / 255f }
         expectedImage.setRGB(3, 1, Color.BLUE, tensorShape, ColorMode.BGR)
@@ -198,7 +198,7 @@ class PreprocessingImageTest {
                 }
             }
             for (i in colorComponents.indices) {
-                set3D(y, x, i, tensorShape[0].toInt(), colorMode.channels, colorComponents[i])
+                set3D(y, x, i, tensorShape[1].toInt(), colorMode.channels, colorComponents[i])
             }
         }
     }