/
ocr_rec.go
131 lines (117 loc) · 3.34 KB
/
ocr_rec.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
package ocr
import (
"log"
"os"
"time"
"github.com/LKKlein/gocv"
)
type TextRecognizer struct {
*PaddleModel
batchNum int
textLen int
shape []int
charType string
labels []string
}
func NewTextRecognizer(modelDir string, args map[string]interface{}) *TextRecognizer {
shapes := []int{3, 32, 320}
if v, ok := args["rec_image_shape"]; ok {
for i, s := range v.([]interface{}) {
shapes[i] = s.(int)
}
}
home, _ := os.UserHomeDir()
labelpath := getString(args, "rec_char_dict_path", home+"/.paddleocr/rec/ppocr_keys_v1.txt")
labels := readLines2StringSlice(labelpath)
if getBool(args, "use_space_char", true) {
labels = append(labels, " ")
}
rec := &TextRecognizer{
PaddleModel: NewPaddleModel(args),
batchNum: getInt(args, "rec_batch_num", 30),
textLen: getInt(args, "max_text_length", 25),
charType: getString(args, "rec_char_type", "ch"),
shape: shapes,
labels: labels,
}
if checkModelExists(modelDir) {
modelDir, _ = downloadModel(home+"/.paddleocr/rec/ch", modelDir)
} else {
log.Panicf("rec model path: %v not exist! Please check!", modelDir)
}
rec.LoadModel(modelDir)
return rec
}
func (rec *TextRecognizer) Run(imgs []gocv.Mat, bboxes [][][]int) []OCRText {
recResult := make([]OCRText, 0, len(imgs))
batch := rec.batchNum
var recTime int64 = 0
c, h, w := rec.shape[0], rec.shape[1], rec.shape[2]
for i := 0; i < len(imgs); i += batch {
j := i + batch
if len(imgs) < j {
j = len(imgs)
}
maxwhratio := 0.0
for k := i; k < j; k++ {
h, w := imgs[k].Rows(), imgs[k].Cols()
ratio := float64(w) / float64(h)
if ratio > maxwhratio {
maxwhratio = ratio
}
}
if rec.charType == "ch" {
w = int(32 * maxwhratio)
}
normimgs := make([]float32, (j-i)*c*h*w)
for k := i; k < j; k++ {
data := crnnPreprocess(imgs[k], rec.shape, []float32{0.5, 0.5, 0.5},
[]float32{0.5, 0.5, 0.5}, 255.0, maxwhratio, rec.charType)
defer imgs[k].Close()
copy(normimgs[(k-i)*c*h*w:], data)
}
st := time.Now()
rec.input.SetValue(normimgs)
rec.input.Reshape([]int32{int32(j - i), int32(c), int32(h), int32(w)})
rec.predictor.SetZeroCopyInput(rec.input)
rec.predictor.ZeroCopyRun()
rec.predictor.GetZeroCopyOutput(rec.outputs[0])
rec.predictor.GetZeroCopyOutput(rec.outputs[1])
recIdxBatch := rec.outputs[0].Value().([][]int64)
recIdxLod := rec.outputs[0].Lod()
predictBatch := rec.outputs[1].Value().([][]float32)
predictLod := rec.outputs[1].Lod()
recTime += int64(time.Since(st).Milliseconds())
for rno := 0; rno < len(recIdxLod)-1; rno++ {
predIdx := make([]int, 0, 2)
for beg := recIdxLod[rno]; beg < recIdxLod[rno+1]; beg++ {
predIdx = append(predIdx, int(recIdxBatch[beg][0]))
}
if len(predIdx) == 0 {
continue
}
words := ""
for n := 0; n < len(predIdx); n++ {
words += rec.labels[predIdx[n]]
}
score := 0.0
count := 0
blankPosition := int(rec.outputs[1].Shape()[1])
for beg := predictLod[rno]; beg < predictLod[rno+1]; beg++ {
argMaxID, maxVal := argmax(predictBatch[beg])
if blankPosition-1-argMaxID > 0 {
score += float64(maxVal)
count++
}
}
score = score / float64(count)
recResult = append(recResult, OCRText{
BBox: bboxes[i+rno],
Text: words,
Score: score,
})
}
}
log.Println("rec num: ", len(recResult), ", rec time elapse: ", recTime, "ms")
return recResult
}