update decode logic
Browse files
app.py
CHANGED
@@ -7,8 +7,9 @@ from PIL import Image
|
|
7 |
from transformers import AutoTokenizer, LayoutLMForQuestionAnswering
|
8 |
from transformers.pipelines.document_question_answering import apply_tesseract
|
9 |
|
10 |
-
|
11 |
-
|
|
|
12 |
OCR = PaddleOCR(
|
13 |
use_angle_cls=True,
|
14 |
lang="en",
|
@@ -56,6 +57,7 @@ def predict(image: Image.Image, question: str, ocr_engine: str):
|
|
56 |
|
57 |
token_ids = TOKENIZER(question)["input_ids"]
|
58 |
token_boxes = [[0] * 4] * (len(token_ids) - 1) + [[1000] * 4]
|
|
|
59 |
|
60 |
token_ids.append(TOKENIZER.sep_token_id)
|
61 |
token_boxes.append([1000] * 4)
|
@@ -74,14 +76,19 @@ def predict(image: Image.Image, question: str, ocr_engine: str):
|
|
74 |
bbox=torch.tensor(token_boxes).unsqueeze(0),
|
75 |
)
|
76 |
|
77 |
-
start_scores = outputs.start_logits.squeeze(0).softmax(-1)
|
78 |
-
end_scores = outputs.end_logits.squeeze(0).softmax(-1)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
|
80 |
-
start_score, start_idx = start_scores.max(-1)
|
81 |
-
end_score, end_idx = end_scores.max(-1)
|
82 |
answer = TOKENIZER.decode(token_ids[start_idx : end_idx + 1])
|
83 |
|
84 |
-
return answer,
|
85 |
|
86 |
|
87 |
gr.Interface(
|
@@ -93,8 +100,7 @@ gr.Interface(
|
|
93 |
],
|
94 |
outputs=[
|
95 |
gr.Textbox(label="Answer"),
|
96 |
-
gr.Number(label="
|
97 |
-
gr.Number(label="End score"),
|
98 |
gr.Image(label="OCR results"),
|
99 |
],
|
100 |
examples=[
|
|
|
7 |
from transformers import AutoTokenizer, LayoutLMForQuestionAnswering
|
8 |
from transformers.pipelines.document_question_answering import apply_tesseract
|
9 |
|
10 |
+
model_tag = "impira/layoutlm-document-qa"
|
11 |
+
MODEL = LayoutLMForQuestionAnswering.from_pretrained(model_tag).eval()
|
12 |
+
TOKENIZER = AutoTokenizer.from_pretrained(model_tag)
|
13 |
OCR = PaddleOCR(
|
14 |
use_angle_cls=True,
|
15 |
lang="en",
|
|
|
57 |
|
58 |
token_ids = TOKENIZER(question)["input_ids"]
|
59 |
token_boxes = [[0] * 4] * (len(token_ids) - 1) + [[1000] * 4]
|
60 |
+
n_question_tokens = len(token_ids)
|
61 |
|
62 |
token_ids.append(TOKENIZER.sep_token_id)
|
63 |
token_boxes.append([1000] * 4)
|
|
|
76 |
bbox=torch.tensor(token_boxes).unsqueeze(0),
|
77 |
)
|
78 |
|
79 |
+
start_scores = outputs.start_logits.squeeze(0).softmax(-1)[n_question_tokens:]
|
80 |
+
end_scores = outputs.end_logits.squeeze(0).softmax(-1)[n_question_tokens:]
|
81 |
+
|
82 |
+
span_scores = start_scores.view(-1, 1) * end_scores.view(1, -1)
|
83 |
+
span_scores = torch.triu(span_scores) # don't allow start < end
|
84 |
+
|
85 |
+
score, indices = span_scores.flatten().max(-1)
|
86 |
+
start_idx = n_question_tokens + indices // span_scores.shape[1]
|
87 |
+
end_idx = n_question_tokens + indices % span_scores.shape[1]
|
88 |
|
|
|
|
|
89 |
answer = TOKENIZER.decode(token_ids[start_idx : end_idx + 1])
|
90 |
|
91 |
+
return answer, score, image_np
|
92 |
|
93 |
|
94 |
gr.Interface(
|
|
|
100 |
],
|
101 |
outputs=[
|
102 |
gr.Textbox(label="Answer"),
|
103 |
+
gr.Number(label="Score"),
|
|
|
104 |
gr.Image(label="OCR results"),
|
105 |
],
|
106 |
examples=[
|