AverageDownsample

- Centered regions are supported for non-discrete widths
- Improved vertical pass and runs much faster now

Author: lessthanoptimal

main/boofcv-ip/src/benchmark/java/boofcv/alg/filter/misc/BenchmarkAverageDownSample.java

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2023, Peter Abeles. All Rights Reserved.
+ * Copyright (c) 2025, Peter Abeles. All Rights Reserved.
  *
  * This file is part of BoofCV (http://boofcv.org).
  *
@@ -24,7 +24,9 @@
 import boofcv.struct.image.GrayF32;
 import boofcv.struct.image.GrayS8;
 import boofcv.struct.image.GrayU8;
+import org.ddogleg.struct.DogArray_F32;
 import org.openjdk.jmh.annotations.*;
+import pabeles.concurrency.GrowArray;
 
 import java.util.Random;
 import java.util.concurrent.TimeUnit;
@@ -44,7 +46,9 @@ public class BenchmarkAverageDownSample {
 
 	GrayU8 inputU8 = new GrayU8(1,1);
 	GrayS8 inputS8 = new GrayS8(1,1);
+	GrayU8 inputF32 = new GrayU8(1,1);
 	GrayU8 out8 = new GrayU8(1,1);
+	GrayF32 tmpF32 = new GrayF32(1,1);
 	GrayF32 outF32 = new GrayF32(1,1);
 
 	@Setup public void setup() {
@@ -53,11 +57,14 @@ public class BenchmarkAverageDownSample {
 
 		inputU8.reshape(size, size);
 		inputS8.reshape(size, size);
+		inputF32.reshape(size, size);
 		out8.reshape(size, size);
+		tmpF32.reshape(size/2, size/2);
 		outF32.reshape(size/2, size/2);
 
 		ImageMiscOps.fillUniform(inputU8,rand,0,200);
 		ImageMiscOps.fillUniform(inputS8,rand,0,200);
+		ImageMiscOps.fillUniform(inputF32,rand,0,200);
 	}
 
 	@Benchmark public void general_8_U8() {
@@ -93,14 +100,29 @@ public class BenchmarkAverageDownSample {
 	}
 
 	@Benchmark public void general_HV_U8() {
-		outF32.reshape(size/2, size);
+		tmpF32.reshape(size/2, size);
 		out8.reshape(size/2, size/2);
+		var workspace = new GrowArray<>(DogArray_F32::new);
+
+		if( BoofConcurrency.USE_CONCURRENT ) {
+			ImplAverageDownSample_MT.horizontal(inputU8, false, tmpF32);
+			ImplAverageDownSample_MT.vertical(tmpF32, false, out8, workspace);
+		} else {
+			ImplAverageDownSample.horizontal(inputU8, false, tmpF32);
+			ImplAverageDownSample.vertical(tmpF32, false, out8, workspace);
+		}
+	}
+
+	@Benchmark public void general_HV_F32() {
+		tmpF32.reshape(size/2, size);
+		out8.reshape(size/2, size/2);
+
 		if( BoofConcurrency.USE_CONCURRENT ) {
-			ImplAverageDownSample_MT.horizontal(inputU8, outF32);
-			ImplAverageDownSample_MT.vertical(outF32, out8);
+			ImplAverageDownSample_MT.horizontal(inputF32, false, tmpF32);
+			ImplAverageDownSample_MT.vertical(tmpF32, false, outF32);
 		} else {
-			ImplAverageDownSample.horizontal(inputU8, outF32);
-			ImplAverageDownSample.vertical(outF32, out8);
+			ImplAverageDownSample.horizontal(inputF32, false, tmpF32);
+			ImplAverageDownSample.vertical(tmpF32, false, outF32);
 		}
 	}
 }