Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,7 +1,10 @@
|
|
1 |
import streamlit as st
|
2 |
from transformers import AutoModelForSeq2SeqLM, T5ForConditionalGeneration, NllbTokenizer, T5Tokenizer
|
|
|
|
|
|
|
|
|
3 |
|
4 |
-
# Initialize models and tokenizers
|
5 |
translation_model_name = 'sarahai/nllb-uzbek-cyrillic-to-russian'
|
6 |
translation_model = AutoModelForSeq2SeqLM.from_pretrained(translation_model_name)
|
7 |
translation_tokenizer = NllbTokenizer.from_pretrained(translation_model_name)
|
@@ -10,10 +13,22 @@ summarization_model_name = 'sarahai/ruT5-base-summarizer'
|
|
10 |
summarization_model = T5ForConditionalGeneration.from_pretrained(summarization_model_name)
|
11 |
summarization_tokenizer = T5Tokenizer.from_pretrained(summarization_model_name)
|
12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
def split_into_chunks(text, tokenizer, max_length=150):
|
14 |
-
# Tokenize the text and get ids
|
15 |
tokens = tokenizer.tokenize(text)
|
16 |
-
# Initialize chunks
|
17 |
chunks = []
|
18 |
current_chunk = []
|
19 |
current_length = 0
|
@@ -24,7 +39,6 @@ def split_into_chunks(text, tokenizer, max_length=150):
|
|
24 |
chunks.append(tokenizer.convert_tokens_to_string(current_chunk))
|
25 |
current_chunk = []
|
26 |
current_length = 0
|
27 |
-
# Add the last chunk if it's not empty
|
28 |
if current_chunk:
|
29 |
chunks.append(tokenizer.convert_tokens_to_string(current_chunk))
|
30 |
return chunks
|
@@ -46,18 +60,28 @@ def summarize(text, model, tokenizer, max_length=250):
|
|
46 |
summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
|
47 |
return summary
|
48 |
|
49 |
-
# Streamlit UI
|
50 |
-
st.title(
|
51 |
-
|
52 |
-
|
53 |
-
if
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
from transformers import AutoModelForSeq2SeqLM, T5ForConditionalGeneration, NllbTokenizer, T5Tokenizer
|
3 |
+
import easyocr
|
4 |
+
from PIL import Image
|
5 |
+
import numpy as np
|
6 |
+
|
7 |
|
|
|
8 |
translation_model_name = 'sarahai/nllb-uzbek-cyrillic-to-russian'
|
9 |
translation_model = AutoModelForSeq2SeqLM.from_pretrained(translation_model_name)
|
10 |
translation_tokenizer = NllbTokenizer.from_pretrained(translation_model_name)
|
|
|
13 |
summarization_model = T5ForConditionalGeneration.from_pretrained(summarization_model_name)
|
14 |
summarization_tokenizer = T5Tokenizer.from_pretrained(summarization_model_name)
|
15 |
|
16 |
+
def extract_text(image_path, lang='uzb_Cyrl'):
|
17 |
+
reader = easyocr.Reader([lang])
|
18 |
+
results = reader.readtext(np.array(image_path))
|
19 |
+
|
20 |
+
all_text = ''
|
21 |
+
confidences = []
|
22 |
+
|
23 |
+
for (bbox, text, prob) in results:
|
24 |
+
all_text += ' ' + text
|
25 |
+
confidences.append(prob)
|
26 |
+
|
27 |
+
final_confidence = sum(confidences) / len(confidences) if confidences else 0
|
28 |
+
return all_text.strip(), final_confidence
|
29 |
+
|
30 |
def split_into_chunks(text, tokenizer, max_length=150):
|
|
|
31 |
tokens = tokenizer.tokenize(text)
|
|
|
32 |
chunks = []
|
33 |
current_chunk = []
|
34 |
current_length = 0
|
|
|
39 |
chunks.append(tokenizer.convert_tokens_to_string(current_chunk))
|
40 |
current_chunk = []
|
41 |
current_length = 0
|
|
|
42 |
if current_chunk:
|
43 |
chunks.append(tokenizer.convert_tokens_to_string(current_chunk))
|
44 |
return chunks
|
|
|
60 |
summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
|
61 |
return summary
|
62 |
|
63 |
+
# Streamlit UI setup
|
64 |
+
st.title('Текстовая обработка изображений, перевод с узбекского на русский и суммаризация')
|
65 |
+
|
66 |
+
uploaded_file = st.file_uploader("Загрузите изображение с узбекским текстом...", type=["jpg", "jpeg", "png"])
|
67 |
+
if uploaded_file is not None:
|
68 |
+
image = Image.open(uploaded_file)
|
69 |
+
st.image(image, caption='Загруженное изображение', use_column_width=True)
|
70 |
+
st.write("Процесс извлечения текста...")
|
71 |
+
extracted_text, confidence = extract_text(image, 'tjk') # Adjust the language code if necessary
|
72 |
+
st.write("Извлеченный текст:")
|
73 |
+
st.text_area("Результат", extracted_text, height=150)
|
74 |
+
st.write(f"Точность распознавания: {confidence*100:.2f}%")
|
75 |
+
|
76 |
+
if st.button("Перевести и суммаризировать"):
|
77 |
+
if extracted_text:
|
78 |
+
with st.spinner('Переводим...'):
|
79 |
+
translated_text = translate(extracted_text, translation_model, translation_tokenizer)
|
80 |
+
st.text_area("Переведенный текст (на русском):", value=translated_text, height=200)
|
81 |
+
|
82 |
+
with st.spinner('Суммаризируем...'):
|
83 |
+
summary_text = summarize(translated_text, summarization_model, summarization_tokenizer, max_length=250)
|
84 |
+
st.text_area("Суммаризация (на русском):", value=summary_text, height=100)
|
85 |
+
else:
|
86 |
+
st.warning("Текст для перевода не найден.")
|
87 |
+
|