-
Notifications
You must be signed in to change notification settings - Fork 0
/
classification.go
329 lines (277 loc) · 8.67 KB
/
classification.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
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
package main
import (
"bytes"
"context"
"fmt"
"image"
"image/jpeg"
"io/ioutil"
"log"
"math"
"regexp"
"strconv"
"strings"
scryfall "github.com/BlueMonday/go-scryfall"
"github.com/BurntSushi/graphics-go/graphics"
"github.com/disintegration/imaging"
"github.com/oliamb/cutter"
"github.com/otiai10/gosseract"
tf "github.com/tensorflow/tensorflow/tensorflow/go"
"github.com/tensorflow/tensorflow/tensorflow/go/op"
"gocv.io/x/gocv"
"gocv.io/x/gocv/contrib"
)
type partialCardData struct {
Name string
TypeLine string
CollectorNumber string
SetSymbol image.Image
}
const (
nameIndex int = iota + 1 // These aren't zero indexed
setSymbolIndex
collectorNumberIndex
typeLineIndex
cardIndex
)
const (
lowestAllowedBoundingBoxProbability = 0.4
minimumOCRWidth = 1500
minimumOCRHeight = 400
)
var (
whitespaceStripper = regexp.MustCompile(`(?m)^\s*$[\r\n]*|[\r\n]+\s+\z`)
collectorNumberNormalizer = regexp.MustCompile(`^0+`)
)
func (d partialCardData) findClosestCard(ctx context.Context, client *scryfall.Client, surf contrib.SURF, setSymbols map[setSymbol]*deferredFile, setSymbolDir string) (scryfall.Card, error) {
if d.Name == "" {
return scryfall.Card{}, nil
}
cardsList, err := client.SearchCards(ctx, d.Name, scryfall.SearchCardsOptions{
Unique: scryfall.UniqueModePrints, // Show everything, including name duplicates
IncludeExtras: true, // Include tokens and things like that
})
if err != nil {
return scryfall.Card{}, err
}
// Get the descriptors for generating SURF keypoints to compare
buf := new(bytes.Buffer)
err = jpeg.Encode(buf, d.SetSymbol, nil)
if err != nil {
return scryfall.Card{}, err
}
refMat := gocv.IMDecode(buf.Bytes(), gocv.IMReadGrayScale)
defer refMat.Close()
_, referenceDescriptors := surf.DetectAndCompute(refMat, gocv.NewMat())
// Find the closest set symbol
var closestDistance float64
var closestIndex int
for i, v := range cardsList.Cards {
collectorNum := collectorNumberNormalizer.ReplaceAllString(strings.Split(d.CollectorNumber, "/")[0], "")
if v.CollectorNumber == collectorNum {
closestIndex = i
break
}
sym := setSymbols[setSymbol{Set: v.Set, Rarity: v.Rarity}]
if sym == nil {
continue
}
buf, err := sym.getBytes(setSymbolDir)
if err != nil {
return scryfall.Card{}, err
}
compMat := gocv.IMDecode(buf, gocv.IMReadGrayScale)
_, comparisonDescriptors := surf.DetectAndCompute(compMat, gocv.NewMat())
resizedRefDesc := gocv.NewMat()
gocv.Resize(referenceDescriptors, &resizedRefDesc, image.Point{X: comparisonDescriptors.Cols(), Y: comparisonDescriptors.Rows()}, 0, 0, gocv.InterpolationLanczos4)
dist := gocv.DifferenceNorm(resizedRefDesc, comparisonDescriptors, gocv.NormL2)
log.Println(dist, setSymbol{Set: v.Set, Rarity: v.Rarity})
if dist > closestDistance || i == 0 {
closestDistance = dist
closestIndex = i
}
}
return cardsList.Cards[closestIndex], nil
}
func (d partialCardData) String() string {
var bound image.Rectangle
if d.SetSymbol == nil {
bound = image.Rectangle{}
} else {
bound = d.SetSymbol.Bounds()
}
return fmt.Sprint("Name:", d.Name, "; TypeLine:", d.TypeLine, "; Collector Number:", d.CollectorNumber, "; Set Symbol Dimensions: ", bound)
}
func inferCroppedSections(imageData []byte, session *tf.Session, graph *tf.Graph) ([]image.Image, []int, error) {
var images []image.Image
var classIndicies []int
tensor, err := makeTensorFromImage(imageData)
if err != nil {
return nil, nil, err
}
// Initialize input/output operations
inputOperation := graph.Operation("image_tensor")
o1 := graph.Operation("detection_boxes")
o2 := graph.Operation("detection_scores")
o3 := graph.Operation("detection_classes")
o4 := graph.Operation("num_detections")
// Execute the graph
output, err := session.Run(
map[tf.Output]*tf.Tensor{
inputOperation.Output(0): tensor,
},
[]tf.Output{
o1.Output(0),
o2.Output(0),
o3.Output(0),
o4.Output(0),
},
nil)
if err != nil {
return nil, nil, err
}
// Actual outputs
probabilities := output[1].Value().([][]float32)[0]
classes := output[2].Value().([][]float32)[0]
boxes := output[0].Value().([][][]float32)[0]
// Pick out the best bounding box of each class
// Both of these are maps so that we can
highestProbabilities := make(map[int]float32) // This is a map so that we can read empty keys
bestIndicies := make(map[int]int) // The key is the class index, the value is the index for the outputs
for i := range probabilities {
if highestProbabilities[int(classes[i])] < probabilities[i] && lowestAllowedBoundingBoxProbability <= probabilities[i] {
highestProbabilities[int(classes[i])] = probabilities[i]
bestIndicies[int(classes[i])] = i
}
}
// Decode the image so we can manipulate it
img, _, err := image.Decode(bytes.NewReader(imageData))
if err != nil {
return nil, nil, err
}
var upsideDown bool
for i := range highestProbabilities {
if i == cardIndex {
continue // Currently unused
}
normalizedY1 := boxes[bestIndicies[i]][0]
normalizedY2 := boxes[bestIndicies[i]][2]
x1 := float32(img.Bounds().Max.X) * boxes[bestIndicies[i]][1]
x2 := float32(img.Bounds().Max.X) * boxes[bestIndicies[i]][3]
y1 := float32(img.Bounds().Max.Y) * normalizedY1
y2 := float32(img.Bounds().Max.Y) * normalizedY2
cropped, err := cutter.Crop(img, cutter.Config{
Width: int(x2 - x1),
Height: int(y2 - y1),
Anchor: image.Point{X: int(x1), Y: int(y1)},
Mode: cutter.TopLeft, // We crop with the anchor at the top left
})
if err != nil {
return nil, nil, err
}
if (normalizedY1+normalizedY2)/2 > 0.5 && i == nameIndex {
upsideDown = true
}
images = append(images, cropped)
classIndicies = append(classIndicies, i)
}
if upsideDown {
for i := range images {
newImage := image.NewRGBA(images[i].Bounds())
graphics.Rotate(newImage, images[i], &graphics.RotateOptions{Angle: math.Pi})
images[i] = newImage
}
log.Println("Upside down.")
}
return images, classIndicies, nil
}
func inferPartialCardData(imageData []byte, session *tf.Session, graph *tf.Graph, tessClient *gosseract.Client) (partialCardData, error) {
crops, classIndicies, err := inferCroppedSections(imageData, session, graph)
if err != nil {
return partialCardData{}, err
}
var inferredData partialCardData
for i, cropped := range crops {
if classIndicies[i] == cardIndex {
continue // Currently unused
} else if classIndicies[i] == setSymbolIndex {
inferredData.SetSymbol = cropped
continue
}
if cropped.Bounds().Dx() < minimumOCRWidth {
cropped = imaging.Resize(cropped, minimumOCRWidth, 0, imaging.Lanczos)
}
if cropped.Bounds().Dy() < minimumOCRHeight {
cropped = imaging.Resize(cropped, 0, minimumOCRHeight, imaging.Lanczos)
}
buf := new(bytes.Buffer)
err = jpeg.Encode(buf, cropped, nil)
if err != nil {
return partialCardData{}, err
}
tessClient.SetImageFromBytes(buf.Bytes())
text, err := tessClient.Text()
if err != nil {
return partialCardData{}, err
}
ioutil.WriteFile(strconv.Itoa(classIndicies[i]), buf.Bytes(), 0644)
text = whitespaceStripper.ReplaceAllString(text, "") // Remove blank lines
switch classIndicies[i] {
case nameIndex:
inferredData.Name = text
case collectorNumberIndex:
inferredData.CollectorNumber = text
case typeLineIndex:
inferredData.TypeLine = text
}
}
return inferredData, nil
}
func newJPEGNormalizationGraph() (graph *tf.Graph, input, output tf.Output, err error) {
s := op.NewScope()
input = op.Placeholder(s, tf.String)
output = op.ExpandDims(s,
op.DecodeJpeg(s, input, op.DecodeJpegChannels(3)),
op.Const(s.SubScope("make_batch"), int32(0)))
graph, err = s.Finalize()
return graph, input, output, err
}
func newDetectionGraph(path string) (*tf.Graph, error) {
// Read the model file
modelBytes, err := ioutil.ReadFile(path)
if err != nil {
return &tf.Graph{}, err
}
// Construct an in-memory graph from the model
graph := tf.NewGraph()
if err := graph.Import(modelBytes, ""); err != nil {
return &tf.Graph{}, err
}
return graph, nil
}
func makeTensorFromImage(imageData []byte) (*tf.Tensor, error) {
// DecodeJpeg uses a scalar String-valued tensor as input.
tensor, err := tf.NewTensor(string(imageData))
if err != nil {
return nil, err
}
// Creates an execution graph to decode and normalize the image
graph, input, output, err := newJPEGNormalizationGraph()
if err != nil {
return nil, err
}
// Execute the graph we just made
session, err := tf.NewSession(graph, nil)
if err != nil {
return nil, err
}
defer session.Close()
normalized, err := session.Run(
map[tf.Output]*tf.Tensor{input: tensor},
[]tf.Output{output},
nil)
if err != nil {
return nil, err
}
return normalized[0], nil
}