Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -29,6 +29,13 @@ processor = LiltForTokenClassification.from_pretrained("SCUT-DLVCLab/lilt-robert
|
|
29 |
model = LayoutLMv3Processor.from_pretrained(
|
30 |
"jinhybr/LiLt-funsd-en"
|
31 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
# load image example
|
34 |
dataset = load_dataset("nielsr/funsd-layoutlmv3", split="test")
|
@@ -38,14 +45,11 @@ image.save("document.png")
|
|
38 |
|
39 |
labels = dataset.features["ner_tags"].feature.names
|
40 |
id2label = {v: k for v, k in enumerate(labels)}
|
41 |
-
label2color = {
|
42 |
-
"question": "blue",
|
43 |
-
"answer": "green",
|
44 |
-
"header": "orange",
|
45 |
-
"other": "violet",
|
46 |
-
}
|
47 |
|
48 |
|
|
|
|
|
|
|
49 |
def unnormalize_box(bbox, width, height):
|
50 |
return [
|
51 |
width * (bbox[0] / 1000),
|
@@ -55,6 +59,18 @@ def unnormalize_box(bbox, width, height):
|
|
55 |
]
|
56 |
|
57 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
def iob_to_label(label):
|
59 |
label = label[2:]
|
60 |
if not label:
|
@@ -62,6 +78,27 @@ def iob_to_label(label):
|
|
62 |
return label
|
63 |
|
64 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
def process_image(image):
|
66 |
width, height = image.size
|
67 |
|
@@ -89,7 +126,9 @@ def process_image(image):
|
|
89 |
if not is_subword[idx]
|
90 |
]
|
91 |
|
92 |
-
|
|
|
|
|
93 |
draw = ImageDraw.Draw(image)
|
94 |
font = ImageFont.load_default()
|
95 |
for prediction, box in zip(true_predictions, true_boxes):
|
@@ -101,6 +140,9 @@ def process_image(image):
|
|
101 |
fill=label2color[predicted_label],
|
102 |
font=font,
|
103 |
)
|
|
|
|
|
|
|
104 |
|
105 |
return image
|
106 |
|
|
|
29 |
model = LayoutLMv3Processor.from_pretrained(
|
30 |
"jinhybr/LiLt-funsd-en"
|
31 |
)
|
32 |
+
####
|
33 |
+
|
34 |
+
|
35 |
+
|
36 |
+
|
37 |
+
|
38 |
+
####
|
39 |
|
40 |
# load image example
|
41 |
dataset = load_dataset("nielsr/funsd-layoutlmv3", split="test")
|
|
|
45 |
|
46 |
labels = dataset.features["ner_tags"].feature.names
|
47 |
id2label = {v: k for v, k in enumerate(labels)}
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
|
49 |
|
50 |
+
|
51 |
+
|
52 |
+
# helper function to unnormalize bboxes for drawing onto the image
|
53 |
def unnormalize_box(bbox, width, height):
|
54 |
return [
|
55 |
width * (bbox[0] / 1000),
|
|
|
59 |
]
|
60 |
|
61 |
|
62 |
+
label2color = {
|
63 |
+
"B-HEADER": "blue",
|
64 |
+
"B-QUESTION": "red",
|
65 |
+
"B-ANSWER": "green",
|
66 |
+
"I-HEADER": "blue",
|
67 |
+
"I-QUESTION": "red",
|
68 |
+
"I-ANSWER": "green",
|
69 |
+
}
|
70 |
+
|
71 |
+
|
72 |
+
|
73 |
+
|
74 |
def iob_to_label(label):
|
75 |
label = label[2:]
|
76 |
if not label:
|
|
|
78 |
return label
|
79 |
|
80 |
|
81 |
+
|
82 |
+
|
83 |
+
# draw results onto the image
|
84 |
+
def draw_boxes(image, boxes, predictions):
|
85 |
+
width, height = image.size
|
86 |
+
normalizes_boxes = [unnormalize_box(box, width, height) for box in boxes]
|
87 |
+
|
88 |
+
# draw predictions over the image
|
89 |
+
draw = ImageDraw.Draw(image)
|
90 |
+
font = ImageFont.load_default()
|
91 |
+
for prediction, box in zip(predictions, normalizes_boxes):
|
92 |
+
if prediction == "O":
|
93 |
+
continue
|
94 |
+
draw.rectangle(box, outline="black")
|
95 |
+
draw.rectangle(box, outline=label2color[prediction])
|
96 |
+
draw.text((box[0] + 10, box[1] - 10), text=prediction, fill=label2color[prediction], font=font)
|
97 |
+
return image
|
98 |
+
|
99 |
+
|
100 |
+
|
101 |
+
|
102 |
def process_image(image):
|
103 |
width, height = image.size
|
104 |
|
|
|
126 |
if not is_subword[idx]
|
127 |
]
|
128 |
|
129 |
+
draw_boxes(image, true_boxes, true_predictions)
|
130 |
+
|
131 |
+
'''' # draw predictions over the image
|
132 |
draw = ImageDraw.Draw(image)
|
133 |
font = ImageFont.load_default()
|
134 |
for prediction, box in zip(true_predictions, true_boxes):
|
|
|
140 |
fill=label2color[predicted_label],
|
141 |
font=font,
|
142 |
)
|
143 |
+
''''
|
144 |
+
|
145 |
+
|
146 |
|
147 |
return image
|
148 |
|