go-rknnlite/example/ppocr/system/system.go

package main

import (
	"flag"
	"fmt"
	"github.com/swdee/go-rknnlite"
	"github.com/swdee/go-rknnlite/postprocess"
	"github.com/swdee/go-rknnlite/preprocess"
	"github.com/swdee/go-rknnlite/render"
	"gocv.io/x/gocv"
	"image"
	"log"
	"math"
	"os"
	"sort"
	"strings"
	"time"
)

func main() {
	// disable logging timestamps
	log.SetFlags(0)

	// read in cli flags
	detectModelFile := flag.String("d", "../../data/models/rk3588/ppocrv4_det-rk3588.rknn", "RKNN compiled model file for OCR Detection")
	recogniseModelFile := flag.String("r", "../../data/models/rk3588/ppocrv4_rec-rk3588.rknn", "RKNN compiled model file for OCR Recognition")
	keysFile := flag.String("k", "../../data/ppocr_keys_v1.txt", "Text file containing OCR character keys")
	imgFile := flag.String("i", "../../data/ppocr-det-test.png", "Image file to run inference on")
	rkPlatform := flag.String("p", "rk3588", "Rockchip CPU Model number [rk3562|rk3566|rk3568|rk3576|rk3582|rk3582|rk3588]")
	flag.Parse()

	err := rknnlite.SetCPUAffinityByPlatform(*rkPlatform, rknnlite.FastCores)

	if err != nil {
		log.Printf("Failed to set CPU Affinity: %v\n", err)
	}

	// check if user specified model file or if default is being used.  if default
	// then pick the default platform model to use.
	if f := flag.Lookup("d"); f != nil && f.Value.String() == f.DefValue && *rkPlatform != "rk3588" {
		*detectModelFile = strings.ReplaceAll(*detectModelFile, "rk3588", *rkPlatform)
	}
	if f := flag.Lookup("r"); f != nil && f.Value.String() == f.DefValue && *rkPlatform != "rk3588" {
		*recogniseModelFile = strings.ReplaceAll(*recogniseModelFile, "rk3588", *rkPlatform)
	}

	// create rknn runtime instance
	detectRt, err := rknnlite.NewRuntimeByPlatform(*rkPlatform, *detectModelFile)

	if err != nil {
		log.Fatal("Error initializing Detect RKNN runtime: ", err)
	}

	recogniseRt, err := rknnlite.NewRuntimeByPlatform(*rkPlatform, *recogniseModelFile)

	if err != nil {
		log.Fatal("Error initializing Recognise RKNN runtime: ", err)
	}

	// set runtime to pass input gocv.Mat's to Inference() function as float32
	// to RKNN backend
	recogniseRt.SetInputTypeFloat32(true)

	// optional querying of model file tensors and SDK version for printing
	// to stdout.  not necessary for production inference code
	err = recogniseRt.Query(os.Stdout)

	if err != nil {
		log.Fatal("Error querying runtime: ", err)
	}

	err = detectRt.Query(os.Stdout)

	if err != nil {
		log.Fatal("Error querying runtime: ", err)
	}

	// load in Model character labels
	modelChars, err := rknnlite.LoadLabels(*keysFile)

	if err != nil {
		log.Fatal("Error loading model OCR character keys: ", err)
	}

	// check that we have as many modelChars as tensor outputs dimension
	if len(modelChars) != int(recogniseRt.OutputAttrs()[0].Dims[2]) {
		log.Fatalf("OCR character keys text input has %d characters and does "+
			"not match the required number in the Model of %d",
			len(modelChars), recogniseRt.OutputAttrs()[0].Dims[2])
	}

	// create PPOCR post processor
	recogniseProcessor := postprocess.NewPPOCRRecognise(postprocess.PPOCRRecogniseParams{
		ModelChars:   modelChars,
		OutputSeqLen: int(recogniseRt.InputAttrs()[0].Dims[2]) / 8, // modelWidth (320/8)
	})

	detectProcessor := postprocess.NewPPOCRDetect(postprocess.PPOCRDetectParams{
		Threshold:    0.3,
		BoxThreshold: 0.6,
		Dilation:     false,
		BoxType:      "poly",
		UnclipRatio:  1.5,
		ScoreMode:    "slow",
		ModelWidth:   int(detectRt.InputAttrs()[0].Dims[2]),
		ModelHeight:  int(detectRt.InputAttrs()[0].Dims[1]),
	})

	// load image
	img := gocv.IMRead(*imgFile, gocv.IMReadColor)

	if img.Empty() {
		log.Fatal("Error reading image from: ", *imgFile)
	}

	// resize image to 480x480 and keep aspect ratio, centered with black letterboxing
	resizedImg := gocv.NewMat()

	resizer := preprocess.NewResizer(img.Cols(), img.Rows(),
		int(detectRt.InputAttrs()[0].Dims[2]), int(detectRt.InputAttrs()[0].Dims[1]),
	)

	resizer.LetterBoxResize(img, &resizedImg, render.Black)

	defer img.Close()
	defer resizedImg.Close()
	defer resizer.Close()

	start := time.Now()

	// perform inference on image file
	outputs, err := detectRt.Inference([]gocv.Mat{resizedImg})

	if err != nil {
		log.Fatal("Runtime inferencing failed with error: ", err)
	}

	// work out scale ratio between source image and resized image
	scaleW := float32(img.Cols()) / float32(resizedImg.Cols())
	scaleH := float32(img.Rows()) / float32(resizedImg.Rows())

	results := detectProcessor.Detect(outputs, scaleW, scaleH)

	// sort results in order from top to bottom and left to right
	SortBoxes(&results)

	endDetect := time.Now() // also start recognise

	// create Mat for cropped region of text
	region := gocv.NewMat()
	defer region.Close()

	for _, result := range results {
		for i, box := range result.Box {
			fmt.Printf("[%d]: [(%d, %d), (%d, %d), (%d, %d), (%d, %d)] %f\n",
				i,
				box.LeftTop.X, box.LeftTop.Y,
				box.RightTop.X, box.RightTop.Y,
				box.RightBottom.X, box.RightBottom.Y,
				box.LeftBottom.X, box.LeftBottom.Y,
				box.Score)

			GetRotateCropImage(img, &region, box)

			// perform text recognition
			recogniseTextBlock(recogniseRt, recogniseProcessor, region,
				int(recogniseRt.InputAttrs()[0].Dims[2]), int(recogniseRt.InputAttrs()[0].Dims[1]))
		}
	}

	endRecognise := time.Now()

	log.Printf("Run speed:\n  Detect processing=%s\n"+
		"  Recognise processing=%s\n"+
		"  Total time=%s\n",
		endDetect.Sub(start).String(),
		endRecognise.Sub(endDetect).String(),
		endRecognise.Sub(start).String(),
	)

	// free outputs allocated in C memory after you have finished post processing
	err = outputs.Free()

	if err != nil {
		log.Fatal("Error freeing Outputs: ", err)
	}

	// close runtime and release resources
	err = detectRt.Close()

	if err != nil {
		log.Fatal("Error closing Detection RKNN runtime: ", err)
	}

	err = recogniseRt.Close()

	if err != nil {
		log.Fatal("Error closing Recognition RKNN runtime: ", err)
	}

	log.Println("done")
}

func recogniseTextBlock(recogniseRt *rknnlite.Runtime,
	recogniseProcessor *postprocess.PPOCRRecognise, img gocv.Mat,
	inWidth, inHeight int) {

	// resize image to 320x48 and keep aspect ratio, centered with black letterboxing
	resizedImg := gocv.NewMat()

	resizer := preprocess.NewResizer(img.Cols(), img.Rows(),
		inWidth, inHeight,
	)

	resizer.LetterBoxResize(img, &resizedImg, render.Black)

	// convert image to float32 in 3 channels
	resizedImg.ConvertTo(&resizedImg, gocv.MatTypeCV32FC3)

	// normalize the image (img - 127.5) / 127.5
	resizedImg.AddFloat(-127.5)
	resizedImg.DivideFloat(127.5)

	defer resizedImg.Close()
	defer resizer.Close()

	// perform inference on image file
	outputs, err := recogniseRt.Inference([]gocv.Mat{resizedImg})

	if err != nil {
		log.Fatal("Runtime inferencing failed with error: ", err)
	}

	results := recogniseProcessor.Recognise(outputs)

	for _, result := range results {
		log.Printf("Recognize result: %s, score=%.2f", result.Text, result.Score)
	}

	// free outputs allocated in C memory after you have finished post processing
	err = outputs.Free()

	if err != nil {
		log.Fatal("Error freeing Outputs: ", err)
	}
}

// CompareBox compares two boxes
func CompareBox(box1, box2 postprocess.PPOCRBox) bool {

	if box1.LeftTop.Y < box2.LeftTop.Y {
		return true
	} else if box1.LeftTop.Y == box2.LeftTop.Y {
		return box1.LeftTop.X < box2.LeftTop.X
	} else {
		return false
	}
}

// SortBoxes sorts the boxes in PPOCRDetectResult and adjusts the order
func SortBoxes(detectResults *[]postprocess.PPOCRDetectResult) {

	for _, result := range *detectResults {

		boxes := result.Box
		sort.Slice(boxes, func(i, j int) bool {
			return CompareBox(boxes[i], boxes[j])
		})

		if len(boxes) == 0 {
			continue
		}

		for i := 0; i < len(boxes)-1; i++ {
			for j := i; j >= 0; j-- {
				if math.Abs(float64(boxes[j+1].LeftTop.Y-boxes[j].LeftTop.Y)) < 10 && (boxes[j+1].LeftTop.X < boxes[j].LeftTop.X) {
					boxes[j], boxes[j+1] = boxes[j+1], boxes[j]
				}
			}
		}
	}
}

// GetRotateCropImage takes the source image and crops it to the bounding box
// and rotates if needed.
func GetRotateCropImage(srcImage gocv.Mat, dstImg *gocv.Mat, box postprocess.PPOCRBox) {

	// Crop the image
	rect := image.Rect(box.LeftTop.X, box.LeftTop.Y, box.RightBottom.X, box.RightBottom.Y)
	region := srcImage.Region(rect)
	imgCrop := region.Clone()
	defer imgCrop.Close()

	// Convert the box points to a slice of image.Point
	points := []image.Point{
		{X: box.LeftTop.X, Y: box.LeftTop.Y},
		{X: box.RightTop.X, Y: box.RightTop.Y},
		{X: box.RightBottom.X, Y: box.RightBottom.Y},
		{X: box.LeftBottom.X, Y: box.LeftBottom.Y},
	}

	// Adjust the points to the coordinates of the cropped image
	left := minInt(
		box.LeftTop.X, box.RightTop.X,
		box.RightBottom.X, box.LeftBottom.X,
	)
	top := minInt(
		box.LeftTop.Y, box.RightTop.Y,
		box.RightBottom.Y, box.LeftBottom.Y,
	)

	// Adjust the points to the cropped region
	for i := range points {
		points[i].X -= left
		points[i].Y -= top
	}

	imgCropWidth := imgCrop.Cols()
	imgCropHeight := imgCrop.Rows()

	// Define the destination points for perspective transformation
	ptsStd := []image.Point{
		{X: 0, Y: 0},
		{X: imgCropWidth, Y: 0},
		{X: imgCropWidth, Y: imgCropHeight},
		{X: 0, Y: imgCropHeight},
	}

	// Get the perspective transform matrix
	srcPoints := gocv.NewPointVectorFromPoints(points)
	dstPoints := gocv.NewPointVectorFromPoints(ptsStd)

	M := gocv.GetPerspectiveTransform(srcPoints, dstPoints)
	defer M.Close()
	srcPoints.Close()
	dstPoints.Close()

	// Apply the warp perspective transformation
	gocv.WarpPerspective(imgCrop, dstImg, M, image.Pt(imgCropWidth, imgCropHeight))

	// Check if the image needs to be transposed and flipped
	if float32(dstImg.Rows()) >= float32(dstImg.Cols())*1.5 {
		srcCopy := gocv.NewMatWithSize(dstImg.Cols(), dstImg.Rows(), dstImg.Type())
		gocv.Transpose(*dstImg, &srcCopy)
		gocv.Flip(srcCopy, &srcCopy, 0)
		*dstImg = srcCopy.Clone()
		srcCopy.Close()
	}
}

// minInt finds the min value in a slice of integers
func minInt(nums ...int) int {

	min := nums[0]

	for _, v := range nums {
		if v < min {
			min = v
		}
	}

	return min
}