feat: new example (#39)

examples/emotions/README.md

@@ -0,0 +1,13 @@
+## Emotion fer+
+
+This is an example of onnx go being able to run the emotion fer+ model
+
+### Howto
+
+- Download an extract the model from [the zoo](https://github.com/onnx/models/tree/master/emotion_ferplus)
+- take a picture of a face: the picture should be 64X64 in gray mode and in png.
+- run `go run main.go -model /path/to/model.onnx -input mypic.png`
+
+
+
+

examples/emotions/main.go

@@ -0,0 +1,124 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"image"
+	"image/png"
+	"io/ioutil"
+	"log"
+	"os"
+	"sort"
+
+	"github.com/owulveryck/onnx-go"
+	"github.com/owulveryck/onnx-go/backend/x/gorgonnx"
+	"github.com/owulveryck/onnx-go/internal/x/images"
+	"gorgonia.org/tensor"
+)
+
+const (
+	height = 64
+	width  = 64
+)
+
+var emotionTable = []string{
+	"neutral",
+	"happiness",
+	"surprise",
+	"sadness",
+	"anger",
+	"disgust",
+	"fear",
+	"contempt",
+}
+
+func main() {
+	model := flag.String("model", "model.onnx", "path to the model file")
+	input := flag.String("input", "file.png", "path to the input file")
+	h := flag.Bool("h", false, "help")
+	flag.Parse()
+	if *h {
+		flag.Usage()
+		os.Exit(0)
+	}
+	for _, f := range []string{*model, *input} {
+		if _, err := os.Stat(f); err != nil && os.IsNotExist(err) {
+			log.Fatalf("%v does not exist", f)
+		}
+	}
+	// Create a backend receiver
+	backend := gorgonnx.NewGraph()
+	// Create a model and set the execution backend
+	m := onnx.NewModel(backend)
+
+	// read the onnx model
+	b, err := ioutil.ReadFile(*model)
+	if err != nil {
+		log.Fatal(err)
+	}
+	// Decode it into the model
+	err = m.UnmarshalBinary(b)
+	if err != nil {
+		log.Fatal(err)
+	}
+	// Set the first input, the number depends of the model
+	// TODO
+	inputFile, err := os.Open(*input)
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer inputFile.Close()
+	img, err := png.Decode(inputFile)
+	if err != nil {
+		log.Fatal(err)
+	}
+	imgGray, ok := img.(*image.Gray)
+	if !ok {
+		log.Fatal("Please give a gray image as input")
+	}
+	inputT := tensor.New(tensor.WithShape(1, 1, height, width), tensor.Of(tensor.Float32))
+	err = images.GrayToBCHW(imgGray, inputT)
+	if err != nil {
+		log.Fatal(err)
+	}
+	m.SetInput(0, inputT)
+	err = backend.Run()
+	if err != nil {
+		log.Fatal(err)
+	}
+	computedOutputT, err := m.GetOutputTensors()
+	if err != nil {
+		log.Fatal(err)
+	}
+	result := classify(computedOutputT[0].Data().([]float32))
+	fmt.Println(result[0].emotion)
+	fmt.Println(result[1].emotion)
+}
+
+type testingT struct{}
+
+func (t *testingT) Errorf(format string, args ...interface{}) {
+	log.Fatalf(format, args...)
+}
+
+func classify(input []float32) emotions {
+	result := make(emotions, len(input))
+	for i := 0; i < len(input); i++ {
+		result[i] = emotion{
+			emotion: emotionTable[i],
+			weight:  input[i],
+		}
+	}
+	sort.Sort(sort.Reverse(result))
+	return result
+}
+
+type emotions []emotion
+type emotion struct {
+	emotion string
+	weight  float32
+}
+
+func (e emotions) Len() int           { return len(e) }
+func (e emotions) Swap(i, j int)      { e[i], e[j] = e[j], e[i] }
+func (e emotions) Less(i, j int) bool { return e[i].weight < e[j].weight }