Added Interval parameter to PhotoInput

This parameter allows us to rescale the input of the models because some
of them need values between [0, 1] and other between [-1, 1].

internal/ai/classify/model.go

@@ -53,10 +53,14 @@ func NewNasnet(assetsPath string, disabled bool) *Model {
 		TFVersion: "1.12.0",
 		Tags:      []string{"photoprism"},
 		Input: &tensorflow.PhotoInput{
-			Name:        "input_1",
-			Height:      224,
-			Width:       224,
-			Channels:    3,
+			Name:     "input_1",
+			Height:   224,
+			Width:    224,
+			Channels: 3,
+			Interval: &tensorflow.Interval{
+				Start: -1,
+				End:   1,
+			},
 			OutputIndex: 0,
 		},
 		Output: &tensorflow.ModelOutput{
@@ -290,5 +294,5 @@ func (m *Model) createTensor(image []byte) (*tf.Tensor, error) {
 		img = imaging.Fill(img, m.meta.Input.Resolution(), m.meta.Input.Resolution(), imaging.Center, imaging.Lanczos)
 	}
 
-	return tensorflow.Image(img, m.meta.Input.Resolution())
+	return tensorflow.Image(img, m.meta.Input)
 }

internal/ai/classify/model_external_test.go

@@ -59,6 +59,12 @@ var modelsInfo = map[string]*tensorflow.ModelInfo{
 		},
 	},
 	"vision-transformer-tensorflow2-vit-b16-classification-v1.tar.gz": &tensorflow.ModelInfo{
+		Input: &tensorflow.PhotoInput{
+			Interval: &tensorflow.Interval{
+				Start: -1.0,
+				End:   1.0,
+			},
+		},
 		Output: &tensorflow.ModelOutput{
 			OutputsLogits: true,
 		},

internal/ai/tensorflow/image.go

@@ -20,11 +20,11 @@ const (
 	Scale = float32(1)
 )
 
-func ImageFromFile(fileName string, resolution int) (*tf.Tensor, error) {
+func ImageFromFile(fileName string, input *PhotoInput) (*tf.Tensor, error) {
 	if img, err := OpenImage(fileName); err != nil {
 		return nil, err
 	} else {
-		return Image(img, resolution)
+		return Image(img, input)
 	}
 }
 
@@ -39,39 +39,39 @@ func OpenImage(fileName string) (image.Image, error) {
 	return img, err
 }
 
-func ImageFromBytes(b []byte, resolution int) (*tf.Tensor, error) {
+func ImageFromBytes(b []byte, input *PhotoInput) (*tf.Tensor, error) {
 	img, _, imgErr := image.Decode(bytes.NewReader(b))
 
 	if imgErr != nil {
 		return nil, imgErr
 	}
 
-	return Image(img, resolution)
+	return Image(img, input)
 }
 
-func Image(img image.Image, resolution int) (tfTensor *tf.Tensor, err error) {
+func Image(img image.Image, input *PhotoInput) (tfTensor *tf.Tensor, err error) {
 	defer func() {
 		if r := recover(); r != nil {
 			err = fmt.Errorf("tensorflow: %s (panic)\nstack: %s", r, debug.Stack())
 		}
 	}()
 
-	if resolution <= 0 {
+	if input.Resolution() <= 0 {
 		return tfTensor, fmt.Errorf("tensorflow: resolution must be larger 0")
 	}
 
 	var tfImage [1][][][3]float32
 
-	for j := 0; j < resolution; j++ {
-		tfImage[0] = append(tfImage[0], make([][3]float32, resolution))
+	for j := 0; j < input.Resolution(); j++ {
+		tfImage[0] = append(tfImage[0], make([][3]float32, input.Resolution()))
 	}
 
-	for i := 0; i < resolution; i++ {
-		for j := 0; j < resolution; j++ {
+	for i := 0; i < input.Resolution(); i++ {
+		for j := 0; j < input.Resolution(); j++ {
 			r, g, b, _ := img.At(i, j).RGBA()
-			tfImage[0][j][i][0] = convertValue(r, 127.5)
-			tfImage[0][j][i][1] = convertValue(g, 127.5)
-			tfImage[0][j][i][2] = convertValue(b, 127.5)
+			tfImage[0][j][i][0] = convertValue(r, input.GetInterval())
+			tfImage[0][j][i][1] = convertValue(g, input.GetInterval())
+			tfImage[0][j][i][2] = convertValue(b, input.GetInterval())
 		}
 	}
 
@@ -136,10 +136,9 @@ func transformImageGraph(imageFormat fs.Type, resolution int) (graph *tf.Graph,
 	return graph, input, output, err
 }
 
-func convertValue(value uint32, mean float32) float32 {
-	if mean == 0 {
-		mean = 127.5
-	}
+func convertValue(value uint32, interval *Interval) float32 {
+	scale := interval.Size() / 255.0
+	offset := interval.Start
 
-	return (float32(value>>8) - mean) / mean
+	return (float32(value>>8))*scale + offset
 }

internal/ai/tensorflow/image_test.go

@@ -10,9 +10,15 @@ import (
 	"github.com/photoprism/photoprism/pkg/fs"
 )
 
+var defaultImageInput = &PhotoInput{
+	Height:   224,
+	Width:    224,
+	Channels: 3,
+}
+
 func TestConvertValue(t *testing.T) {
-	result := convertValue(uint32(98765432), 127.5)
-	assert.Equal(t, float32(3024.898), result)
+	result := convertValue(uint32(98765432), &Interval{Start: -1, End: 1})
+	assert.Equal(t, float32(3024.8982), result)
 }
 
 func TestImageFromBytes(t *testing.T) {
@@ -26,7 +32,7 @@ func TestImageFromBytes(t *testing.T) {
 			t.Fatal(err)
 		}
 
-		result, err := ImageFromBytes(imageBuffer, 224)
+		result, err := ImageFromBytes(imageBuffer, defaultImageInput)
 		assert.Equal(t, tensorflow.DataType(0x1), result.DataType())
 		assert.Equal(t, int64(1), result.Shape()[0])
 		assert.Equal(t, int64(224), result.Shape()[2])
@@ -34,7 +40,7 @@ func TestImageFromBytes(t *testing.T) {
 	t.Run("Document", func(t *testing.T) {
 		imageBuffer, err := os.ReadFile(examplesPath + "/Random.docx")
 		assert.Nil(t, err)
-		result, err := ImageFromBytes(imageBuffer, 224)
+		result, err := ImageFromBytes(imageBuffer, defaultImageInput)
 
 		assert.Empty(t, result)
 		assert.EqualError(t, err, "image: unknown format")

internal/ai/tensorflow/info.go

@@ -11,13 +11,33 @@ import (
 	"google.golang.org/protobuf/proto"
 )
 
+// Interval of allowed values
+type Interval struct {
+	Start float32 `yaml:"Start,omitempty" json:"start,omitempty"`
+	End   float32 `yaml:"End,omitempty" json:"end,omitempty"`
+}
+
+// The size of the interval
+func (i Interval) Size() float32 {
+	return i.End - i.Start
+}
+
+// The standard interval returned by decodeImage is [0, 1]
+func StandardInterval() *Interval {
+	return &Interval{
+		Start: 0.0,
+		End:   1.0,
+	}
+}
+
 // Input description for a photo input for a model
 type PhotoInput struct {
-	Name        string `yaml:"Name,omitempty" json:"name,omitempty"`
-	OutputIndex int    `yaml:"Index,omitempty" json:"index,omitempty"`
-	Height      int64  `yaml:"Height,omitempty" json:"height,omitempty"`
-	Width       int64  `yaml:"Width,omitempty" json:"width,omitempty"`
-	Channels    int64  `yaml:"Channels,omitempty" json:"channels,omitempty"`
+	Name        string    `yaml:"Name,omitempty" json:"name,omitempty"`
+	Interval    *Interval `yaml:"Interval,omitempty" json:"interval,omitempty"`
+	OutputIndex int       `yaml:"Index,omitempty" json:"index,omitempty"`
+	Height      int64     `yaml:"Height,omitempty" json:"height,omitempty"`
+	Width       int64     `yaml:"Width,omitempty" json:"width,omitempty"`
+	Channels    int64     `yaml:"Channels,omitempty" json:"channels,omitempty"`
 }
 
 // When dimensions are not defined, it means the model accepts any size of
@@ -37,12 +57,25 @@ func (p *PhotoInput) SetResolution(resolution int) {
 	p.Width = int64(resolution)
 }
 
+// Get the interval or the default one
+func (p PhotoInput) GetInterval() *Interval {
+	if p.Interval == nil {
+		return StandardInterval()
+	} else {
+		return p.Interval
+	}
+}
+
 // Merge other input with this.
 func (p *PhotoInput) Merge(other *PhotoInput) {
 	if p.Name == "" {
 		p.Name = other.Name
 	}
 
+	if p.Interval == nil && other.Interval != nil {
+		p.Interval = other.Interval
+	}
+
 	if p.OutputIndex == 0 {
 		p.OutputIndex = other.OutputIndex
 	}

internal/ai/vision/models.go

@@ -16,9 +16,13 @@ var (
 			TFVersion: "1.12.0",
 			Tags:      []string{"photoprism"},
 			Input: &tensorflow.PhotoInput{
-				Name:        "input_1",
-				Height:      224,
-				Width:       224,
+				Name:   "input_1",
+				Height: 224,
+				Width:  224,
+				Interval: &tensorflow.Interval{
+					Start: -1.0,
+					End:   1.0,
+				},
 				Channels:    3,
 				OutputIndex: 0,
 			},