Fix scaling functions (#385)

This commit corrects two small errors that causes the improper
results reported in #312.

- A copy paste typo (most likely) which sets the Dy component
  of the image rectangle to Dx.
- The tmp array in fastBoxSampling was not properly
  reset because the last argument in memset should be
  the number of bytes to fill in and not the number of elements.

.github/workflows/ci.yaml

@@ -17,9 +17,9 @@ jobs:
     name: Linux Go ${{ matrix.go }}
     steps:
       - name: Checkout
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
       - name: Setup Go
-        uses: actions/setup-go@v2
+        uses: actions/setup-go@v3
         with:
           go-version: ${{ matrix.go }}
       - name: Install dependencies
@@ -43,9 +43,9 @@ jobs:
     name: Darwin Go ${{ matrix.go }}
     steps:
       - name: Checkout
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
       - name: Setup Go
-        uses: actions/setup-go@v2
+        uses: actions/setup-go@v3
         with:
           go-version: ${{ matrix.go }}
       - name: Install dependencies
@@ -62,9 +62,9 @@ jobs:
     name: Check Licenses
     steps:
       - name: Checkout
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
       - name: Setup Go
-        uses: actions/setup-go@v2
+        uses: actions/setup-go@v3
         with:
           go-version: '1.16'
       - name: Installing go-licenses

.github/workflows/renovate-go-mod-fix.yaml

@@ -9,7 +9,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: checkout
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
         with:
           fetch-depth: 2
       - name: fix

go.mod

@@ -10,6 +10,6 @@ require (
 	github.com/pion/logging v0.2.2
 	github.com/pion/rtp v1.7.4
 	github.com/pion/webrtc/v3 v3.1.10
-	golang.org/x/image v0.0.0-20210628002857-a66eb6448b8d
+	golang.org/x/image v0.0.0-20220302094943-723b81ca9867
 	gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c // indirect
 )

go.sum

@@ -100,8 +100,8 @@ golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPh
 golang.org/x/crypto v0.0.0-20210322153248-0c34fe9e7dc2/go.mod h1:T9bdIzuCu7OtxOm1hfPfRQxPLYneinmdGuTeoZ9dtd4=
 golang.org/x/crypto v0.0.0-20210921155107-089bfa567519 h1:7I4JAnoQBe7ZtJcBaYHi5UtiO8tQHbUSXxL+pnGRANg=
 golang.org/x/crypto v0.0.0-20210921155107-089bfa567519/go.mod h1:GvvjBRRGRdwPK5ydBHafDWAxML/pGHZbMvKqRZ5+Abc=
-golang.org/x/image v0.0.0-20210628002857-a66eb6448b8d h1:RNPAfi2nHY7C2srAV8A49jpsYr0ADedCk1wq6fTMTvs=
+golang.org/x/image v0.0.0-20220302094943-723b81ca9867 h1:TcHcE0vrmgzNH1v3ppjcMGbhG5+9fMuvOmUYwNEF4q4=
-golang.org/x/image v0.0.0-20210628002857-a66eb6448b8d/go.mod h1:023OzeP/+EPmXeapQh35lcL3II3LrY8Ic+EFFKVhULM=
+golang.org/x/image v0.0.0-20220302094943-723b81ca9867/go.mod h1:023OzeP/+EPmXeapQh35lcL3II3LrY8Ic+EFFKVhULM=
 golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/net v0.0.0-20180906233101-161cd47e91fd/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=

pkg/codec/openh264/bridge.cpp

@@ -88,9 +88,8 @@ Slice enc_encode(Encoder *e, Frame f, int *eresult) {
   pic.iPicWidth = f.width;
   pic.iPicHeight = f.height;
   pic.iColorFormat = videoFormatI420;
-  // We always received I420 format
+  pic.iStride[0] = f.ystride;
-  pic.iStride[0] = pic.iPicWidth;
+  pic.iStride[1] = pic.iStride[2] = f.cstride;
-  pic.iStride[1] = pic.iStride[2] = pic.iPicWidth / 2;
   pic.pData[0] = (unsigned char *)f.y;
   pic.pData[1] = (unsigned char *)f.u;
   pic.pData[2] = (unsigned char *)f.v;

pkg/codec/openh264/bridge.hpp

@@ -12,6 +12,8 @@ typedef struct Slice {
 
 typedef struct Frame {
   void *y, *u, *v;
+  int ystride;
+  int cstride;
   int height;
   int width;
 } Frame;

pkg/codec/openh264/openh264.go

@@ -65,11 +65,13 @@ func (e *encoder) Read() ([]byte, func(), error) {
 	bounds := yuvImg.Bounds()
 	var rv C.int
 	s := C.enc_encode(e.engine, C.Frame{
-		y:      unsafe.Pointer(&yuvImg.Y[0]),
+		y:       unsafe.Pointer(&yuvImg.Y[0]),
-		u:      unsafe.Pointer(&yuvImg.Cb[0]),
+		u:       unsafe.Pointer(&yuvImg.Cb[0]),
-		v:      unsafe.Pointer(&yuvImg.Cr[0]),
+		v:       unsafe.Pointer(&yuvImg.Cr[0]),
-		height: C.int(bounds.Max.Y - bounds.Min.Y),
+		ystride: C.int(yuvImg.YStride),
-		width:  C.int(bounds.Max.X - bounds.Min.X),
+		cstride: C.int(yuvImg.CStride),
+		height:  C.int(bounds.Max.Y - bounds.Min.Y),
+		width:   C.int(bounds.Max.X - bounds.Min.X),
 	}, &rv)
 	if err := errResult(rv); err != nil {
 		return nil, func() {}, fmt.Errorf("failed in encoding: %v", err)

pkg/io/video/scale.go

@@ -109,7 +109,7 @@ func Scale(width, height int, scaler Scaler) TransformFunc {
 			yDy := rect.Dy()
 			cRect := fixedRect(rect, i1.SubsampleRatio)
 			cDx := cRect.Dx()
-			cDy := cRect.Dx()
+			cDy := cRect.Dy()
 			yLen := yDx * yDy
 			cLen := cDx * cDy
 			if len(imgDst.Y) < yLen {

pkg/io/video/scale_test.go

@@ -36,6 +36,32 @@ func TestScale(t *testing.T) {
 				Rect:   image.Rect(0, 0, 2, 2),
 			},
 		},
+		"RGBASameSize": {
+			src: &image.RGBA{
+				Pix: []uint8{
+					// R     G     B     A |   R     G     B     A |   R     G     B     A |   R     G     B     A
+					0x80, 0x00, 0x00, 0xFF, 0x80, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF,
+					0x80, 0x00, 0x00, 0xFF, 0x80, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF,
+					0x00, 0x40, 0x00, 0xFF, 0x00, 0x40, 0x00, 0xFF, 0x00, 0x00, 0x60, 0xFF, 0x00, 0x00, 0x60, 0xFF,
+					0x00, 0x40, 0x00, 0xFF, 0x00, 0x40, 0x00, 0xFF, 0x00, 0x00, 0x60, 0xFF, 0x00, 0x00, 0x60, 0xFF,
+				},
+				Stride: 16,
+				Rect:   image.Rect(0, 0, 4, 4),
+			},
+			width:  4,
+			height: 4,
+			expected: &image.RGBA{
+				Pix: []uint8{
+					// R     G     B     A |   R     G     B     A |   R     G     B     A |   R     G     B     A
+					0x80, 0x00, 0x00, 0xFF, 0x80, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF,
+					0x80, 0x00, 0x00, 0xFF, 0x80, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0xFF,
+					0x00, 0x40, 0x00, 0xFF, 0x00, 0x40, 0x00, 0xFF, 0x00, 0x00, 0x60, 0xFF, 0x00, 0x00, 0x60, 0xFF,
+					0x00, 0x40, 0x00, 0xFF, 0x00, 0x40, 0x00, 0xFF, 0x00, 0x00, 0x60, 0xFF, 0x00, 0x00, 0x60, 0xFF,
+				},
+				Stride: 16,
+				Rect:   image.Rect(0, 0, 4, 4),
+			},
+		},
 		"I444": {
 			src: &image.YCbCr{
 				SubsampleRatio: image.YCbCrSubsampleRatio444,
@@ -91,6 +117,70 @@ func TestScale(t *testing.T) {
 				Rect:    image.Rect(0, 0, 3, 3),
 			},
 		},
+		"I444SameSize": {
+			src: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio444,
+				Y: []uint8{
+					0xF0, 0xF0, 0x00, 0x00, 0x00, 0x00,
+					0xF0, 0xF0, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40,
+					0x00, 0x00, 0x80, 0x80, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x00, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x20, 0x80, 0x80, 0x80, 0x80,
+					0x20, 0x20, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80, 0xC0, 0xC0,
+					0x80, 0x80, 0x80, 0x80, 0xC0, 0xC0,
+					0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+				},
+				Cr: []uint8{
+					0xE0, 0xE0, 0x80, 0x80, 0x80, 0x80,
+					0xE0, 0xE0, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x40, 0x40, 0x80, 0x80,
+					0x80, 0x80, 0x40, 0x40, 0x80, 0x80,
+				},
+				YStride: 6,
+				CStride: 6,
+				Rect:    image.Rect(0, 0, 6, 6),
+			},
+			width:  6,
+			height: 6,
+			expected: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio444,
+				Y: []uint8{
+					0xF0, 0xF0, 0x00, 0x00, 0x00, 0x00,
+					0xF0, 0xF0, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40,
+					0x00, 0x00, 0x80, 0x80, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x00, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x20, 0x80, 0x80, 0x80, 0x80,
+					0x20, 0x20, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80, 0xC0, 0xC0,
+					0x80, 0x80, 0x80, 0x80, 0xC0, 0xC0,
+					0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+				},
+				Cr: []uint8{
+					0xE0, 0xE0, 0x80, 0x80, 0x80, 0x80,
+					0xE0, 0xE0, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x40, 0x40, 0x80, 0x80,
+					0x80, 0x80, 0x40, 0x40, 0x80, 0x80,
+				},
+				YStride: 6,
+				CStride: 6,
+				Rect:    image.Rect(0, 0, 6, 6),
+			},
+		},
 		"I422": {
 			src: &image.YCbCr{
 				SubsampleRatio: image.YCbCrSubsampleRatio422,
@@ -155,6 +245,82 @@ func TestScale(t *testing.T) {
 				Rect:    image.Rect(0, 0, 4, 4),
 			},
 		},
+		"I422SameSize": {
+			src: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio422,
+				Y: []uint8{
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00,
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x20, 0x80, 0x80,
+					0x20, 0x20, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0xE0, 0xE0,
+					0x80, 0x80, 0xE0, 0xE0,
+				},
+				Cr: []uint8{
+					0xE0, 0xE0, 0x80, 0x80,
+					0xE0, 0xE0, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0xF0, 0xF0, 0x40, 0x40,
+					0xF0, 0xF0, 0x40, 0x40,
+				},
+				YStride: 8,
+				CStride: 4,
+				Rect:    image.Rect(0, 0, 8, 8),
+			},
+			width:  8,
+			height: 8,
+			expected: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio422,
+				Y: []uint8{
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00,
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x20, 0x80, 0x80,
+					0x20, 0x20, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0xE0, 0xE0,
+					0x80, 0x80, 0xE0, 0xE0,
+				},
+				Cr: []uint8{
+					0xE0, 0xE0, 0x80, 0x80,
+					0xE0, 0xE0, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0x80, 0x80, 0x80, 0x80,
+					0xF0, 0xF0, 0x40, 0x40,
+					0xF0, 0xF0, 0x40, 0x40,
+				},
+				YStride: 8,
+				CStride: 4,
+				Rect:    image.Rect(0, 0, 8, 8),
+			},
+		},
 		"I420": {
 			src: &image.YCbCr{
 				SubsampleRatio: image.YCbCrSubsampleRatio420,
@@ -207,6 +373,118 @@ func TestScale(t *testing.T) {
 				Rect:    image.Rect(0, 0, 4, 4),
 			},
 		},
+		"I420SameSize": {
+			src: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio420,
+				Y: []uint8{
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00,
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x20, 0x80, 0x80,
+					0x20, 0x20, 0x80, 0x80,
+					0x80, 0x80, 0xE0, 0xE0,
+					0x80, 0x80, 0xE0, 0xE0,
+				},
+				Cr: []uint8{
+					0xE0, 0xE0, 0x80, 0x80,
+					0xE0, 0xE0, 0x80, 0x80,
+					0xF0, 0xF0, 0x40, 0x40,
+					0xF0, 0xF0, 0x40, 0x40,
+				},
+				YStride: 8,
+				CStride: 4,
+				Rect:    image.Rect(0, 0, 8, 8),
+			},
+			width:  8,
+			height: 8,
+			expected: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio420,
+				Y: []uint8{
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00,
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x20, 0x80, 0x80,
+					0x20, 0x20, 0x80, 0x80,
+					0x80, 0x80, 0xE0, 0xE0,
+					0x80, 0x80, 0xE0, 0xE0,
+				},
+				Cr: []uint8{
+					0xE0, 0xE0, 0x80, 0x80,
+					0xE0, 0xE0, 0x80, 0x80,
+					0xF0, 0xF0, 0x40, 0x40,
+					0xF0, 0xF0, 0x40, 0x40,
+				},
+				YStride: 8,
+				CStride: 4,
+				Rect:    image.Rect(0, 0, 8, 8),
+			},
+		},
+		"I420NonSquareImage": {
+			src: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio420,
+				Y: []uint8{
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0xF0, 0xF0, 0x10, 0x10,
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0xF0, 0xF0, 0x10, 0x10,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00, 0x30, 0x30, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00, 0x30, 0x30, 0x00, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x20, 0x80, 0x80, 0x50, 0x50,
+					0x20, 0x20, 0x80, 0x80, 0x50, 0x50,
+					0x80, 0x80, 0xE0, 0xE0, 0x30, 0x30,
+					0x80, 0x80, 0xE0, 0xE0, 0x30, 0x30,
+				},
+				Cr: []uint8{
+					0xE0, 0xE0, 0x80, 0x80, 0xB0, 0xB0,
+					0xE0, 0xE0, 0x80, 0x80, 0xB0, 0xB0,
+					0xF0, 0xF0, 0x40, 0x40, 0xC0, 0xC0,
+					0xF0, 0xF0, 0x40, 0x40, 0xC0, 0xC0,
+				},
+				YStride: 12,
+				CStride: 6,
+				Rect:    image.Rect(0, 0, 12, 8),
+			},
+			width:  6,
+			height: 4,
+			expected: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio420,
+				Y: []uint8{
+					0xF0, 0x10, 0x00, 0x00, 0xF0, 0x10,
+					0x00, 0x00, 0x40, 0x00, 0x40, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x80, 0x30, 0x00, 0x30, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x80, 0x50,
+					0x80, 0xE0, 0x30,
+				},
+				Cr: []uint8{
+					0xE0, 0x80, 0xB0,
+					0xF0, 0x40, 0xC0,
+				},
+				YStride: 6,
+				CStride: 3,
+				Rect:    image.Rect(0, 0, 6, 4),
+			},
+		},
 	}
 	for name, algo := range scalerTestAlgos {
 		algo := algo
@@ -247,6 +525,100 @@ func TestScale(t *testing.T) {
 	}
 }
 
+func TestScaleFastBoxSampling(t *testing.T) {
+	cases := map[string]struct {
+		src           image.Image
+		width, height int
+		expected      image.Image
+	}{
+		"I420NonSquareImage": {
+			src: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio420,
+				Y: []uint8{
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0xF0, 0xF0, 0x10, 0x10,
+					0xF0, 0xF0, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0xF0, 0xF0, 0x10, 0x10,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00, 0x40, 0x40, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00, 0x30, 0x30, 0x00, 0x00,
+					0x00, 0x00, 0x80, 0x80, 0x30, 0x30, 0x00, 0x00, 0x30, 0x30, 0x00, 0x00,
+				},
+				Cb: []uint8{
+					0x20, 0x20, 0x80, 0x80, 0x50, 0x50,
+					0x20, 0x20, 0x80, 0x80, 0x50, 0x50,
+					0x80, 0x80, 0xE0, 0xE0, 0x30, 0x30,
+					0x80, 0x80, 0xE0, 0xE0, 0x30, 0x30,
+				},
+				Cr: []uint8{
+					0xE0, 0xE0, 0x80, 0x80, 0xB0, 0xB0,
+					0xE0, 0xE0, 0x80, 0x80, 0xB0, 0xB0,
+					0xF0, 0xF0, 0x40, 0x40, 0xC0, 0xC0,
+					0xF0, 0xF0, 0x40, 0x40, 0xC0, 0xC0,
+				},
+				YStride: 12,
+				CStride: 6,
+				Rect:    image.Rect(0, 0, 12, 8),
+			},
+			width:  6,
+			height: 4,
+			expected: &image.YCbCr{
+				SubsampleRatio: image.YCbCrSubsampleRatio420,
+				Y: []uint8{
+					0xF0, 0x80, 0x08, 0x00, 0x78, 0x80,
+					0x08, 0x00, 0x20, 0x20, 0x20, 0x20,
+					0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+					0x00, 0x40, 0x58, 0x18, 0x18, 0x18,
+				},
+				Cb: []uint8{
+					0x20, 0x50, 0x68,
+					0x68, 0xB0, 0x88,
+				},
+				Cr: []uint8{
+					0xE0, 0xB0, 0x98,
+					0xD0, 0x98, 0x80,
+				},
+				YStride: 6,
+				CStride: 3,
+				Rect:    image.Rect(0, 0, 6, 4),
+			},
+		},
+	}
+
+	for name, c := range cases {
+		c := c
+		t.Run(name, func(t *testing.T) {
+			trans := Scale(c.width, c.height, ScalerFastBoxSampling)
+			r := trans(ReaderFunc(func() (image.Image, func(), error) {
+				return c.src, func() {}, nil
+			}))
+			for i := 0; i < 4; i++ {
+				out, _, err := r.Read()
+				if err != nil {
+					t.Fatalf("Unexpected error: %v", err)
+				}
+				if !reflect.DeepEqual(c.expected, out) {
+					t.Errorf("Expected output image:\n%v\ngot:\n%v\nrepeat: %d", c.expected, out, i)
+				}
+				// Destroy output contents
+				switch v := out.(type) {
+				case *image.RGBA:
+					v.Stride = 10
+					v.Pix = v.Pix[:1]
+					v.Rect.Max.X = 1
+				case *image.YCbCr:
+					v.YStride = 10
+					v.CStride = 100
+					v.Y = v.Y[:1]
+					v.Cb = v.Cb[:2]
+					v.Cr = v.Cr[:1]
+					v.Rect.Max.X = 1
+				}
+			}
+		})
+	}
+}
+
 func BenchmarkScale(b *testing.B) {
 	for name, algo := range scalerBenchAlgos {
 		algo := algo

pkg/io/video/scaler_cgo.c

@@ -29,7 +29,7 @@ void fastBoxSampling(
     const int sw, const int sh, const int sstride,
     uint32_t* tmp)
 {
-  memset(tmp, 0, dw * dh * ch);
+  memset(tmp, 0, dw * dh * ch * sizeof(tmp[0]));
 
   for (int sy = 0; sy < sh; sy++)
   {

pkg/io/video/scaleycrcb.go

@@ -30,10 +30,18 @@ func (p *rgbLikeYCbCr) At(x, y int) color.Color {
 }
 
 func (p *rgbLikeYCbCr) Set(x, y int, c color.Color) {
-	rgb := c.(*color.RGBA64)
+	switch v := c.(type) {
-	p.y.SetGray(x, y, color.Gray{uint8(rgb.R / 0x100)})
+	case color.RGBA:
-	if (image.Point{x, y}.In(p.cb.Rect)) {
+		p.y.SetGray(x, y, color.Gray{v.R})
-		p.cb.SetGray(x, y, color.Gray{uint8(rgb.G / 0x100)})
+		if (image.Point{x, y}.In(p.cb.Rect)) {
-		p.cr.SetGray(x, y, color.Gray{uint8(rgb.B / 0x100)})
+			p.cb.SetGray(x, y, color.Gray{v.G})
+			p.cr.SetGray(x, y, color.Gray{v.B})
+		}
+	case *color.RGBA64:
+		p.y.SetGray(x, y, color.Gray{uint8(v.R / 0x100)})
+		if (image.Point{x, y}.In(p.cb.Rect)) {
+			p.cb.SetGray(x, y, color.Gray{uint8(v.G / 0x100)})
+			p.cr.SetGray(x, y, color.Gray{uint8(v.B / 0x100)})
+		}
 	}
 }