aico commited on
Commit
f4d5676
1 Parent(s): e2823b0

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +74 -8
app.py CHANGED
@@ -7,26 +7,92 @@ import cv2
7
  processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
8
  model = VisionEncoderDecoderModel.from_pretrained("aico/TrOCR-MNIST")
9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
 
11
- def process_image(image):
12
 
 
 
 
 
13
  #print(np.shape(image))
14
  #print(image)
 
 
15
  #rint(image.astype('uint8'))
16
  #cv2.imwrite("image.png",image.astype('uint8'),(28, 28))
17
- img = Image.fromarray(image.astype('uint8')).convert("RGB")
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
18
  #img = Image.open("image.png").convert("RGB")
19
- print(img)
 
20
  # prepare image
21
- pixel_values = processor(img, return_tensors="pt").pixel_values
22
 
23
  # generate (no beam search)
24
- generated_ids = model.generate(pixel_values)
25
 
26
  # decode
27
- generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
28
-
29
- return generated_text
30
 
31
  title = "Interactive demo: Single Digits MNIST"
32
  description = "Aico - University Utrecht"
 
7
  processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
8
  model = VisionEncoderDecoderModel.from_pretrained("aico/TrOCR-MNIST")
9
 
10
+ def _group_rectangles(rec):
11
+ """
12
+ Uion intersecting rectangles.
13
+ Args:
14
+ rec - list of rectangles in form [x, y, w, h]
15
+ Return:
16
+ list of grouped ractangles
17
+ """
18
+ tested = [False for i in range(len(rec))]
19
+ final = []
20
+ i = 0
21
+ while i < len(rec):
22
+ if not tested[i]:
23
+ j = i+1
24
+ while j < len(rec):
25
+ if not tested[j] and intersect_area(rec[i], rec[j]):
26
+ rec[i] = union(rec[i], rec[j])
27
+ tested[j] = True
28
+ j = i
29
+ j += 1
30
+ final += [rec[i]]
31
+ i += 1
32
 
33
+ return final
34
 
35
+ def process_image(image):
36
+ bounding_boxes = []
37
+ generated_text_list = []
38
+ #boundingBoxes_2 = []
39
  #print(np.shape(image))
40
  #print(image)
41
+ #dim = (28,28)
42
+ #resized = cv2.resize(image, dim, interpolation = cv2.INTER_AREA)
43
  #rint(image.astype('uint8'))
44
  #cv2.imwrite("image.png",image.astype('uint8'),(28, 28))
45
+ #mask = np.zeros(np.shape(image), dtype=np.uint8)
46
+ thresh = cv2.threshold(image, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
47
+ #gray = cv2.cvtColor(thresh, cv2.COLOR_BGR2GRAY)
48
+
49
+ cnts = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
50
+ cnts = cnts[0] if len(cnts) == 2 else cnts[1]
51
+ (cnts, _) = contours.sort_contours(cnts, method="left-to-right")
52
+ dim = (28, 28)
53
+ for c in cnts:
54
+ area = cv2.contourArea(c)
55
+ #print(area)
56
+ #if area < 120:
57
+ bounding_boxes.append(cv2.boundingRect(c))
58
+ print("for loop bb: ",bounding_boxes)
59
+
60
+ boundingBoxes_filter = [i for i in bounding_boxes if i != (0 , 0, 128, 128)]
61
+
62
+ boundingBoxes = _group_rectangles(boundingBoxes_filter)
63
+ #print(boundingBoxes)
64
+ #
65
+ #print(boundingBoxes_2)
66
+ for (x, y, w, h) in boundingBoxes:
67
+ print(x,y,w,h)
68
+ ROI = thresh[y:y+h, x:x+w]
69
+ ROI2 = cv2.bitwise_not(ROI)
70
+ borderoutput = cv2.copyMakeBorder(ROI2, 30, 30, 30, 30, cv2.BORDER_CONSTANT, value=[0, 0, 0])
71
+
72
+ resized = cv2.resize(borderoutput, dim, interpolation = cv2.INTER_AREA)
73
+ cv2.imwrite('ROI_{}.png'.format(x), resized)
74
+ #imageinv = cv2.bitwise_not(resized)
75
+ img = Image.fromarray(resized.astype('uint8')).convert("RGB")
76
+
77
+ pixel_values = processor(img, return_tensors="pt").pixel_values
78
+ generated_ids = model.generate(pixel_values)
79
+ generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
80
+ print(generated_text)
81
+ generated_text_list.append(generated_text)
82
+ #img = Image.fromarray(image.astype('uint8')).convert("RGB")
83
  #img = Image.open("image.png").convert("RGB")
84
+ #print(img)
85
+
86
  # prepare image
87
+ #pixel_values = processor(img, return_tensors="pt").pixel_values
88
 
89
  # generate (no beam search)
90
+ #generated_ids = model.generate(pixel_values)
91
 
92
  # decode
93
+ #generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
94
+ return ''.join(generated_text_list)
95
+ #return generated_text
96
 
97
  title = "Interactive demo: Single Digits MNIST"
98
  description = "Aico - University Utrecht"