Spaces:
Sleeping
Sleeping
viboognesh-doaz
commited on
Commit
·
b075822
1
Parent(s):
da07179
updated code
Browse files- app.py +35 -232
- awsfunctions.py +164 -0
- pdf_processing.py +148 -0
- qdrant_mm_db_pipeline/.lock +0 -1
- qdrant_mm_db_pipeline/collection/image_collection_pipeline/storage.sqlite +0 -0
- qdrant_mm_db_pipeline/collection/text_collection_pipeline/storage.sqlite +0 -3
- qdrant_mm_db_pipeline/meta.json +0 -1
- qdrant_mm_db_pipeline_tmp0uzyg0nb_construction_pdf/.lock +0 -1
- qdrant_mm_db_pipeline_tmp0uzyg0nb_construction_pdf/collection/image_collection_pipeline_tmp0uzyg0nb_construction_pdf/storage.sqlite +0 -3
- qdrant_mm_db_pipeline_tmp0uzyg0nb_construction_pdf/collection/text_collection_pipeline_tmp0uzyg0nb_construction_pdf/storage.sqlite +0 -3
- qdrant_mm_db_pipeline_tmp0uzyg0nb_construction_pdf/meta.json +0 -1
- requirements.txt +2 -1
- retrieve_and_display.py +59 -0
app.py
CHANGED
@@ -1,218 +1,13 @@
|
|
1 |
import streamlit as st
|
2 |
-
import os
|
3 |
-
from PyPDF2 import PdfReader
|
4 |
-
import pymupdf
|
5 |
-
import numpy as np
|
6 |
-
import cv2
|
7 |
-
import shutil
|
8 |
-
import imageio
|
9 |
-
from PIL import Image
|
10 |
-
import imagehash
|
11 |
-
import matplotlib.pyplot as plt
|
12 |
-
from llama_index.core.indices import MultiModalVectorStoreIndex
|
13 |
-
from llama_index.vector_stores.qdrant import QdrantVectorStore
|
14 |
-
from llama_index.core import SimpleDirectoryReader, StorageContext
|
15 |
-
import qdrant_client
|
16 |
-
from llama_index.core import PromptTemplate
|
17 |
-
from llama_index.core.query_engine import SimpleMultiModalQueryEngine
|
18 |
-
from llama_index.llms.openai import OpenAI
|
19 |
-
from llama_index.core import load_index_from_storage, get_response_synthesizer
|
20 |
-
import tempfile
|
21 |
-
from qdrant_client import QdrantClient, models
|
22 |
-
import getpass
|
23 |
|
24 |
-
|
25 |
-
|
26 |
-
# To connect to the same event-loop,
|
27 |
-
# allows async events to run on notebook
|
28 |
-
|
29 |
-
# import nest_asyncio
|
30 |
-
|
31 |
-
# nest_asyncio.apply()
|
32 |
|
33 |
from dotenv import load_dotenv
|
34 |
load_dotenv()
|
35 |
|
36 |
-
|
37 |
-
|
38 |
-
reader = PdfReader(pdf_path)
|
39 |
-
full_text = ''
|
40 |
-
for page in reader.pages:
|
41 |
-
text = page.extract_text()
|
42 |
-
full_text += text
|
43 |
-
return full_text
|
44 |
-
|
45 |
-
def extract_images_from_pdf(pdf_path, img_save_path):
|
46 |
-
doc = pymupdf.open(pdf_path)
|
47 |
-
for page in doc:
|
48 |
-
img_number = 0
|
49 |
-
for block in page.get_text("dict")["blocks"]:
|
50 |
-
if block['type'] == 1:
|
51 |
-
name = os.path.join(img_save_path, f"img{page.number}-{img_number}.{block['ext']}")
|
52 |
-
out = open(name, "wb")
|
53 |
-
out.write(block["image"])
|
54 |
-
out.close()
|
55 |
-
img_number += 1
|
56 |
-
|
57 |
-
def is_empty(img_path):
|
58 |
-
image = cv2.imread(img_path, 0)
|
59 |
-
std_dev = np.std(image)
|
60 |
-
return std_dev < 1
|
61 |
-
|
62 |
-
def move_images(source_folder, dest_folder):
|
63 |
-
image_files = [f for f in os.listdir(source_folder)
|
64 |
-
if f.lower().endswith(('.jpg', '.jpeg', '.png', '.gif'))]
|
65 |
-
os.makedirs(dest_folder, exist_ok=True)
|
66 |
-
moved_count = 0
|
67 |
-
for file in image_files:
|
68 |
-
src_path = os.path.join(source_folder, file)
|
69 |
-
if not is_empty(src_path):
|
70 |
-
shutil.move(src_path, os.path.join(dest_folder, file))
|
71 |
-
moved_count += 1
|
72 |
-
return moved_count
|
73 |
-
|
74 |
-
def remove_low_size_images(data_path):
|
75 |
-
images_list = os.listdir(data_path)
|
76 |
-
low_size_photo_list = []
|
77 |
-
for one_image in images_list:
|
78 |
-
image_path = os.path.join(data_path, one_image)
|
79 |
-
try:
|
80 |
-
pic = imageio.imread(image_path)
|
81 |
-
size = pic.size
|
82 |
-
if size < 100:
|
83 |
-
low_size_photo_list.append(one_image)
|
84 |
-
except:
|
85 |
-
pass
|
86 |
-
for one_image in low_size_photo_list[1:]:
|
87 |
-
os.remove(os.path.join(data_path, one_image))
|
88 |
-
|
89 |
-
def calc_diff(img1 , img2) :
|
90 |
-
i1 = Image.open(img1)
|
91 |
-
i2 = Image.open(img2)
|
92 |
-
h1 = imagehash.phash(i1)
|
93 |
-
h2 = imagehash.phash(i2)
|
94 |
-
return h1 - h2
|
95 |
-
|
96 |
-
def remove_duplicate_images(data_path) :
|
97 |
-
image_files = os.listdir(data_path)
|
98 |
-
only_images = []
|
99 |
-
for one_image in image_files :
|
100 |
-
if one_image.endswith('jpeg') or one_image.endswith('png') or one_image.endswith('jpg') :
|
101 |
-
only_images.append(one_image)
|
102 |
-
only_images1 = sorted(only_images)
|
103 |
-
for one_image in only_images1 :
|
104 |
-
for another_image in only_images1 :
|
105 |
-
try :
|
106 |
-
if one_image == another_image :
|
107 |
-
continue
|
108 |
-
else :
|
109 |
-
diff = calc_diff(os.path.join(data_path ,one_image) , os.path.join(data_path ,another_image))
|
110 |
-
if diff ==0 :
|
111 |
-
os.remove(os.path.join(data_path , another_image))
|
112 |
-
except Exception as e:
|
113 |
-
print(e)
|
114 |
-
pass
|
115 |
-
# from langchain_chroma import Chroma
|
116 |
-
# import chromadb
|
117 |
-
def initialize_qdrant(temp_dir , file_name , user):
|
118 |
-
client = qdrant_client.QdrantClient(path=f"qdrant_mm_db_pipeline_{user}_{file_name}")
|
119 |
-
# client = qdrant_client.QdrantClient(url = "http://localhost:2452")
|
120 |
-
# client = qdrant_client.QdrantClient(url="4b0af7be-d5b3-47ac-b215-128ebd6aa495.europe-west3-0.gcp.cloud.qdrant.io:6333", api_key="CO1sNGLmC6R_Q45qSIUxBSX8sxwHud4MCm4as_GTI-vzQqdUs-bXqw",)
|
121 |
-
# client = qdrant_client.AsyncQdrantClient(location = ":memory:")
|
122 |
-
|
123 |
-
if "vectordatabase" not in st.session_state or not st.session_state.vectordatabase:
|
124 |
-
|
125 |
-
# text_store = client.create_collection(f"text_collection_pipeline_{user}_{file_name}" )
|
126 |
-
# image_store = client.create_collection(f"image_collection_pipeline_{user}_{file_name}" )
|
127 |
-
|
128 |
-
|
129 |
-
text_store = QdrantVectorStore( client = client , collection_name=f"text_collection_pipeline_{user}_{file_name}" )
|
130 |
-
image_store = QdrantVectorStore(client = client , collection_name=f"image_collection_pipeline_{user}_{file_name}")
|
131 |
-
storage_context = StorageContext.from_defaults(vector_store=text_store, image_store=image_store)
|
132 |
-
documents = SimpleDirectoryReader(os.path.join(temp_dir, f"my_own_data_{user}_{file_name}")).load_data()
|
133 |
-
index = MultiModalVectorStoreIndex.from_documents(documents, storage_context=storage_context)
|
134 |
-
|
135 |
-
st.session_state.vectordatabase = index
|
136 |
-
else :
|
137 |
-
index = st.session_state.vectordatabase
|
138 |
-
retriever_engine = index.as_retriever(similarity_top_k=1, image_similarity_top_k=1)
|
139 |
-
return retriever_engine
|
140 |
-
|
141 |
-
def plot_images(image_paths):
|
142 |
-
images_shown = 0
|
143 |
-
plt.figure(figsize=(16, 9))
|
144 |
-
for img_path in image_paths:
|
145 |
-
if os.path.isfile(img_path):
|
146 |
-
image = Image.open(img_path)
|
147 |
-
plt.subplot(2, 3, images_shown + 1)
|
148 |
-
plt.imshow(image)
|
149 |
-
plt.xticks([])
|
150 |
-
plt.yticks([])
|
151 |
-
images_shown += 1
|
152 |
-
if images_shown >= 6:
|
153 |
-
break
|
154 |
-
|
155 |
-
def retrieve_and_query(query, retriever_engine):
|
156 |
-
retrieval_results = retriever_engine.retrieve(query)
|
157 |
-
|
158 |
-
qa_tmpl_str = (
|
159 |
-
"Context information is below.\n"
|
160 |
-
"---------------------\n"
|
161 |
-
"{context_str}\n"
|
162 |
-
"---------------------\n"
|
163 |
-
"Given the context information , "
|
164 |
-
"answer the query in detail.\n"
|
165 |
-
"Query: {query_str}\n"
|
166 |
-
"Answer: "
|
167 |
-
)
|
168 |
-
qa_tmpl = PromptTemplate(qa_tmpl_str)
|
169 |
-
|
170 |
-
llm = OpenAI(model="gpt-4o", temperature=0)
|
171 |
-
response_synthesizer = get_response_synthesizer(response_mode="refine", text_qa_template=qa_tmpl, llm=llm)
|
172 |
-
|
173 |
-
response = response_synthesizer.synthesize(query, nodes=retrieval_results)
|
174 |
-
|
175 |
-
retrieved_image_path_list = []
|
176 |
-
for node in retrieval_results:
|
177 |
-
if (node.metadata['file_type'] == 'image/jpeg') or (node.metadata['file_type'] == 'image/png'):
|
178 |
-
if node.score > 0.25:
|
179 |
-
retrieved_image_path_list.append(node.metadata['file_path'])
|
180 |
-
|
181 |
-
return response, retrieved_image_path_list
|
182 |
-
#tmpnimvp35m , tmpnimvp35m , tmpydpissmv
|
183 |
-
def process_pdf(pdf_file):
|
184 |
-
temp_dir = tempfile.TemporaryDirectory()
|
185 |
-
unique_folder_name = temp_dir.name.split('/')[-1]
|
186 |
-
temp_pdf_path = os.path.join(temp_dir.name, pdf_file.name)
|
187 |
-
with open(temp_pdf_path, "wb") as f:
|
188 |
-
f.write(pdf_file.getvalue())
|
189 |
-
|
190 |
-
data_path = os.path.join(temp_dir.name, f"my_own_data_{unique_folder_name}_{os.path.splitext(pdf_file.name)[0]}")
|
191 |
-
os.makedirs(data_path , exist_ok=True)
|
192 |
-
img_save_path = os.path.join(temp_dir.name, f"extracted_images_{unique_folder_name}_{os.path.splitext(pdf_file.name)[0]}")
|
193 |
-
os.makedirs(img_save_path , exist_ok=True)
|
194 |
-
|
195 |
-
extracted_text = extract_text_from_pdf(temp_pdf_path)
|
196 |
-
with open(os.path.join(data_path, "content.txt"), "w") as file:
|
197 |
-
file.write(extracted_text)
|
198 |
-
|
199 |
-
extract_images_from_pdf(temp_pdf_path, img_save_path)
|
200 |
-
moved_count = move_images(img_save_path, data_path)
|
201 |
-
remove_low_size_images(data_path)
|
202 |
-
remove_duplicate_images(data_path)
|
203 |
-
retriever_engine = initialize_qdrant(temp_dir.name , os.path.splitext(pdf_file.name)[0] , unique_folder_name)
|
204 |
-
|
205 |
-
return temp_dir, retriever_engine
|
206 |
-
|
207 |
-
def main():
|
208 |
-
st.title("PDF Vector Database Query Tool")
|
209 |
-
st.markdown("This tool creates a vector database from a PDF and allows you to query it.")
|
210 |
-
|
211 |
-
if "retriever_engine" not in st.session_state:
|
212 |
-
st.session_state.retriever_engine = None
|
213 |
-
if "vectordatabase" not in st.session_state:
|
214 |
-
st.session_state.vectordatabase = None
|
215 |
-
|
216 |
uploaded_file = st.file_uploader("Choose a PDF file", type="pdf")
|
217 |
if uploaded_file is None:
|
218 |
st.info("Please upload a PDF file.")
|
@@ -220,34 +15,42 @@ def main():
|
|
220 |
st.info(f"Uploaded PDF: {uploaded_file.name}")
|
221 |
if st.button("Process PDF"):
|
222 |
with st.spinner("Processing PDF..."):
|
223 |
-
|
224 |
-
|
225 |
st.success("PDF processed successfully!")
|
226 |
|
|
|
227 |
if st.session_state.retriever_engine :
|
228 |
-
|
229 |
-
|
230 |
-
|
231 |
-
|
232 |
-
|
233 |
-
|
|
|
|
|
|
|
|
|
|
|
234 |
|
235 |
-
|
236 |
-
|
237 |
-
|
238 |
-
|
239 |
-
|
240 |
-
|
241 |
-
|
242 |
-
|
243 |
-
|
244 |
-
|
|
|
|
|
|
|
|
|
245 |
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
except Exception as e:
|
250 |
-
st.error(f"An error occurred: {e}")
|
251 |
|
252 |
if __name__ == "__main__":
|
|
|
253 |
main()
|
|
|
|
1 |
import streamlit as st
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
+
from pdf_processing import process_pdf
|
4 |
+
from retrieve_and_display import retrieve_and_query, plot_images
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
from dotenv import load_dotenv
|
7 |
load_dotenv()
|
8 |
|
9 |
+
def upload_file():
|
10 |
+
st.title("Upload File to chat with file")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
uploaded_file = st.file_uploader("Choose a PDF file", type="pdf")
|
12 |
if uploaded_file is None:
|
13 |
st.info("Please upload a PDF file.")
|
|
|
15 |
st.info(f"Uploaded PDF: {uploaded_file.name}")
|
16 |
if st.button("Process PDF"):
|
17 |
with st.spinner("Processing PDF..."):
|
18 |
+
st.session_state.retriever_engine = process_pdf(uploaded_file)
|
|
|
19 |
st.success("PDF processed successfully!")
|
20 |
|
21 |
+
def ask_question():
|
22 |
if st.session_state.retriever_engine :
|
23 |
+
if user_question := st.chat_input("Ask a question"):
|
24 |
+
with st.spinner("Retrieving information..."):
|
25 |
+
response, retrieved_image_path_list = retrieve_and_query(user_question, st.session_state.retriever_engine)
|
26 |
+
print(retrieved_image_path_list)
|
27 |
+
st.write("Retrieved Context:")
|
28 |
+
for node in response.source_nodes:
|
29 |
+
st.code(node.node.get_text())
|
30 |
+
|
31 |
+
st.write("\nRetrieved Images:")
|
32 |
+
plot_images(retrieved_image_path_list)
|
33 |
+
st.pyplot()
|
34 |
|
35 |
+
st.write("\nFinal Answer:")
|
36 |
+
st.code(response.response)
|
37 |
+
else:
|
38 |
+
st.title("Upload File to chat with file")
|
39 |
+
|
40 |
+
|
41 |
+
|
42 |
+
def main():
|
43 |
+
if "retriever_engine" not in st.session_state:
|
44 |
+
st.session_state.retriever_engine = None
|
45 |
+
page_names_to_funcs = {
|
46 |
+
"Upload File": upload_file,
|
47 |
+
"Chat": ask_question
|
48 |
+
}
|
49 |
|
50 |
+
demo_name = st.sidebar.selectbox("PDF Query Tool", page_names_to_funcs.keys())
|
51 |
+
page_names_to_funcs[demo_name]()
|
|
|
|
|
|
|
52 |
|
53 |
if __name__ == "__main__":
|
54 |
+
# login_page()
|
55 |
main()
|
56 |
+
|
awsfunctions.py
ADDED
@@ -0,0 +1,164 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
import boto3
|
3 |
+
from botocore.exceptions import ClientError, NoCredentialsError
|
4 |
+
import os
|
5 |
+
|
6 |
+
def upload_folder_to_s3(local_dir, prefix=''):
|
7 |
+
s3_bucket = os.getenv("AWS_BUCKET_NAME")
|
8 |
+
s3_client = boto3.client('s3')
|
9 |
+
|
10 |
+
for root, dirs, files in os.walk(local_dir):
|
11 |
+
for dir in dirs:
|
12 |
+
dir_path = os.path.join(root, dir)
|
13 |
+
relative_path = os.path.relpath(dir_path, local_dir)
|
14 |
+
|
15 |
+
# Create the directory in S3 if it doesn't exist
|
16 |
+
try:
|
17 |
+
s3_client.put_object(Bucket=s3_bucket, Key=f"{prefix}{relative_path}/")
|
18 |
+
except ClientError as e:
|
19 |
+
if e.response['Error']['Code'] == '404':
|
20 |
+
continue # Directory already exists
|
21 |
+
else:
|
22 |
+
raise
|
23 |
+
|
24 |
+
for file in files:
|
25 |
+
file_path = os.path.join(root, file)
|
26 |
+
relative_path = os.path.relpath(file_path, local_dir)
|
27 |
+
|
28 |
+
try:
|
29 |
+
s3_client.upload_file(file_path, s3_bucket, f"{prefix}{relative_path}")
|
30 |
+
print(f"Uploaded: {file_path} -> s3://{s3_bucket}/{prefix}{relative_path}")
|
31 |
+
except Exception as e:
|
32 |
+
raise e
|
33 |
+
# print(f"Error uploading {file_path}: {e}")
|
34 |
+
|
35 |
+
def check_file_exists_in_s3(file_path):
|
36 |
+
bucket_name = os.getenv("AWS_BUCKET_NAME")
|
37 |
+
s3_client = boto3.client('s3')
|
38 |
+
|
39 |
+
try:
|
40 |
+
s3_client.head_object(Bucket=bucket_name, Key=file_path)
|
41 |
+
return True
|
42 |
+
except ClientError as e:
|
43 |
+
if e.response['Error']['Code'] == '404':
|
44 |
+
return False
|
45 |
+
else:
|
46 |
+
raise e
|
47 |
+
|
48 |
+
def download_files_from_s3(local_folder, file_path_list):
|
49 |
+
s3 = boto3.client('s3')
|
50 |
+
bucket_name = os.getenv("AWS_BUCKET_NAME")
|
51 |
+
folder_prefix = ''
|
52 |
+
|
53 |
+
try:
|
54 |
+
# List objects in the S3 bucket
|
55 |
+
paginator = s3.get_paginator('list_objects_v2')
|
56 |
+
page_iterator = paginator.paginate(Bucket=bucket_name, Prefix=folder_prefix)
|
57 |
+
|
58 |
+
# Download filtered files
|
59 |
+
for page in page_iterator:
|
60 |
+
for obj in page.get('Contents', []):
|
61 |
+
key = obj['Key']
|
62 |
+
|
63 |
+
# Apply file filter if specified
|
64 |
+
if key not in file_path_list:
|
65 |
+
continue
|
66 |
+
|
67 |
+
# Construct local file path
|
68 |
+
local_path = os.path.join(local_folder, key)
|
69 |
+
os.makedirs(os.path.dirname(local_path), exist_ok=True)
|
70 |
+
try:
|
71 |
+
print(f"Downloading: {key} -> {local_path}")
|
72 |
+
s3.download_file(bucket_name, key, local_path)
|
73 |
+
print(f"Downloaded: {local_path}")
|
74 |
+
except Exception as e:
|
75 |
+
print(f"Error downloading {key}: {e}")
|
76 |
+
|
77 |
+
except NoCredentialsError:
|
78 |
+
print("No AWS credentials found.")
|
79 |
+
except Exception as e:
|
80 |
+
print(f"An error occurred: {e}")
|
81 |
+
|
82 |
+
def download_folder_from_s3(local_folder, aws_folder_prefix):
|
83 |
+
s3 = boto3.client('s3')
|
84 |
+
bucket_name = os.getenv("AWS_BUCKET_NAME")
|
85 |
+
folder_prefix = aws_folder_prefix
|
86 |
+
|
87 |
+
try:
|
88 |
+
# List objects in the S3 bucket
|
89 |
+
paginator = s3.get_paginator('list_objects_v2')
|
90 |
+
page_iterator = paginator.paginate(Bucket=bucket_name, Prefix=folder_prefix)
|
91 |
+
|
92 |
+
# Download filtered files
|
93 |
+
for page in page_iterator:
|
94 |
+
for obj in page.get('Contents', []):
|
95 |
+
key = obj['Key']
|
96 |
+
|
97 |
+
# Construct local file path
|
98 |
+
local_path = os.path.join(local_folder, key)
|
99 |
+
os.makedirs(os.path.dirname(local_path), exist_ok=True)
|
100 |
+
try:
|
101 |
+
print(f"Downloading: {key} -> {local_path}")
|
102 |
+
s3.download_file(bucket_name, key, local_path)
|
103 |
+
print(f"Downloaded: {local_path}")
|
104 |
+
except Exception as e:
|
105 |
+
print(f"Error downloading {key}: {e}")
|
106 |
+
raise e
|
107 |
+
|
108 |
+
except NoCredentialsError:
|
109 |
+
print("No AWS credentials found.")
|
110 |
+
except Exception as e:
|
111 |
+
print(f"An error occurred: {e}")
|
112 |
+
|
113 |
+
def delete_s3_folder(folder_path):
|
114 |
+
bucket_name = os.getenv("AWS_BUCKET_NAME")
|
115 |
+
s3_client = boto3.client('s3')
|
116 |
+
|
117 |
+
try:
|
118 |
+
# List objects in the S3 bucket
|
119 |
+
paginator = s3_client.get_paginator('list_objects_v2')
|
120 |
+
page_iterator = paginator.paginate(Bucket=bucket_name, Prefix=folder_path)
|
121 |
+
|
122 |
+
# Delete objects within the folder_path
|
123 |
+
delete_keys = {'Objects': []}
|
124 |
+
for page in page_iterator:
|
125 |
+
for obj in page.get('Contents', []):
|
126 |
+
key = obj['Key']
|
127 |
+
|
128 |
+
# Construct the full key for deletion
|
129 |
+
delete_key = {'Key': key}
|
130 |
+
delete_keys['Objects'].append(delete_key)
|
131 |
+
|
132 |
+
print(f"Deleting: {key}")
|
133 |
+
|
134 |
+
# Perform batch delete operation
|
135 |
+
if len(delete_keys['Objects']) > 0:
|
136 |
+
s3_client.delete_objects(Bucket=bucket_name, Delete=delete_keys)
|
137 |
+
print(f"Deleted {len(delete_keys['Objects'])} objects in folder '{folder_path}'")
|
138 |
+
else:
|
139 |
+
print(f"No objects found in folder '{folder_path}'")
|
140 |
+
|
141 |
+
except ClientError as e:
|
142 |
+
print(f"An error occurred: {e}")
|
143 |
+
|
144 |
+
def list_s3_objects(prefix=''):
|
145 |
+
bucket_name = os.getenv("AWS_BUCKET_NAME")
|
146 |
+
s3_client = boto3.client('s3')
|
147 |
+
|
148 |
+
try:
|
149 |
+
paginator = s3_client.get_paginator('list_objects_v2')
|
150 |
+
page_iterator = paginator.paginate(Bucket=bucket_name, Prefix=prefix)
|
151 |
+
for page in page_iterator:
|
152 |
+
for obj in page.get('Contents', []):
|
153 |
+
print(f"Key: {obj['Key']}")
|
154 |
+
print(f"Size: {obj['Size']} bytes")
|
155 |
+
print(f"Last Modified: {obj['LastModified']}")
|
156 |
+
print(f"ETag: {obj['ETag']}")
|
157 |
+
print(f"File Extension: {os.path.splitext(obj['Key'])[-1]}")
|
158 |
+
print("---")
|
159 |
+
|
160 |
+
except ClientError as e:
|
161 |
+
print(f"An error occurred: {e}")
|
162 |
+
|
163 |
+
|
164 |
+
|
pdf_processing.py
ADDED
@@ -0,0 +1,148 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from PyPDF2 import PdfReader
|
2 |
+
import pymupdf
|
3 |
+
import numpy as np
|
4 |
+
import cv2
|
5 |
+
import shutil
|
6 |
+
import imageio
|
7 |
+
from PIL import Image
|
8 |
+
import imagehash
|
9 |
+
import tempfile
|
10 |
+
import os
|
11 |
+
from llama_index.core.indices import MultiModalVectorStoreIndex
|
12 |
+
from llama_index.vector_stores.qdrant import QdrantVectorStore
|
13 |
+
from llama_index.core import SimpleDirectoryReader, StorageContext
|
14 |
+
from awsfunctions import upload_folder_to_s3, check_file_exists_in_s3, download_folder_from_s3
|
15 |
+
import qdrant_client
|
16 |
+
import streamlit as st
|
17 |
+
|
18 |
+
def extract_text_from_pdf(pdf_path):
|
19 |
+
reader = PdfReader(pdf_path)
|
20 |
+
full_text = ''
|
21 |
+
for page in reader.pages:
|
22 |
+
text = page.extract_text()
|
23 |
+
full_text += text
|
24 |
+
return full_text
|
25 |
+
|
26 |
+
def extract_images_from_pdf(pdf_path, img_save_path):
|
27 |
+
doc = pymupdf.open(pdf_path)
|
28 |
+
for page in doc:
|
29 |
+
img_number = 0
|
30 |
+
for block in page.get_text("dict")["blocks"]:
|
31 |
+
if block['type'] == 1:
|
32 |
+
name = os.path.join(img_save_path, f"img{page.number}-{img_number}.{block['ext']}")
|
33 |
+
out = open(name, "wb")
|
34 |
+
out.write(block["image"])
|
35 |
+
out.close()
|
36 |
+
img_number += 1
|
37 |
+
|
38 |
+
def is_empty(img_path):
|
39 |
+
image = cv2.imread(img_path, 0)
|
40 |
+
std_dev = np.std(image)
|
41 |
+
return std_dev < 1
|
42 |
+
|
43 |
+
def move_images(source_folder, dest_folder):
|
44 |
+
image_files = [f for f in os.listdir(source_folder)
|
45 |
+
if f.lower().endswith(('.jpg', '.jpeg', '.png', '.gif'))]
|
46 |
+
os.makedirs(dest_folder, exist_ok=True)
|
47 |
+
moved_count = 0
|
48 |
+
for file in image_files:
|
49 |
+
src_path = os.path.join(source_folder, file)
|
50 |
+
if not is_empty(src_path):
|
51 |
+
shutil.move(src_path, os.path.join(dest_folder, file))
|
52 |
+
moved_count += 1
|
53 |
+
return moved_count
|
54 |
+
|
55 |
+
def remove_low_size_images(data_path):
|
56 |
+
images_list = os.listdir(data_path)
|
57 |
+
low_size_photo_list = []
|
58 |
+
for one_image in images_list:
|
59 |
+
image_path = os.path.join(data_path, one_image)
|
60 |
+
try:
|
61 |
+
pic = imageio.imread(image_path)
|
62 |
+
size = pic.size
|
63 |
+
if size < 100:
|
64 |
+
low_size_photo_list.append(one_image)
|
65 |
+
except:
|
66 |
+
pass
|
67 |
+
for one_image in low_size_photo_list[1:]:
|
68 |
+
os.remove(os.path.join(data_path, one_image))
|
69 |
+
|
70 |
+
def calc_diff(img1 , img2) :
|
71 |
+
i1 = Image.open(img1)
|
72 |
+
i2 = Image.open(img2)
|
73 |
+
h1 = imagehash.phash(i1)
|
74 |
+
h2 = imagehash.phash(i2)
|
75 |
+
return h1 - h2
|
76 |
+
|
77 |
+
def remove_duplicate_images(data_path) :
|
78 |
+
image_files = os.listdir(data_path)
|
79 |
+
only_images = []
|
80 |
+
for one_image in image_files :
|
81 |
+
if one_image.endswith('jpeg') or one_image.endswith('png') or one_image.endswith('jpg') :
|
82 |
+
only_images.append(one_image)
|
83 |
+
only_images1 = sorted(only_images)
|
84 |
+
for one_image in only_images1 :
|
85 |
+
for another_image in only_images1 :
|
86 |
+
try :
|
87 |
+
if one_image == another_image :
|
88 |
+
continue
|
89 |
+
else :
|
90 |
+
diff = calc_diff(os.path.join(data_path ,one_image) , os.path.join(data_path ,another_image))
|
91 |
+
if diff ==0 :
|
92 |
+
os.remove(os.path.join(data_path , another_image))
|
93 |
+
except Exception as e:
|
94 |
+
print(e)
|
95 |
+
pass
|
96 |
+
# from langchain_chroma import Chroma
|
97 |
+
# import chromadb
|
98 |
+
def initialize_qdrant(temp_dir , aws_prefix):
|
99 |
+
client = qdrant_client.QdrantClient(path=os.path.join(temp_dir, "qdrant"))
|
100 |
+
text_store = QdrantVectorStore( client = client , collection_name=f"text_collection" )
|
101 |
+
image_store = QdrantVectorStore(client = client , collection_name=f"image_collection")
|
102 |
+
storage_context = StorageContext.from_defaults(vector_store=text_store, image_store=image_store)
|
103 |
+
documents = SimpleDirectoryReader(os.path.join(temp_dir, f"data")).load_data()
|
104 |
+
for doc in documents:
|
105 |
+
doc.metadata["file_path"] = os.path.join(aws_prefix, os.path.relpath(doc.metadata["file_path"], temp_dir))
|
106 |
+
index = MultiModalVectorStoreIndex.from_documents(documents, storage_context=storage_context)
|
107 |
+
retriever_engine = index.as_retriever(similarity_top_k=1, image_similarity_top_k=1)
|
108 |
+
return retriever_engine
|
109 |
+
|
110 |
+
def process_pdf(pdf_file):
|
111 |
+
username = "ptchecker"
|
112 |
+
aws_prefix_path = os.path.join(os.getenv("FOLDER_PREFIX"), username, "FILES", os.path.splitext(pdf_file.name)[0])
|
113 |
+
if check_file_exists_in_s3(os.path.join(aws_prefix_path, pdf_file.name)):
|
114 |
+
temp_dir = tempfile.mkdtemp()
|
115 |
+
download_folder_from_s3(local_folder=temp_dir, aws_folder_prefix=os.path.join(aws_prefix_path, "qdrant"))
|
116 |
+
client = qdrant_client.QdrantClient(path=os.path.join(temp_dir))
|
117 |
+
image_store = QdrantVectorStore(client = client , collection_name=f"image_collection")
|
118 |
+
text_store = QdrantVectorStore(client = client , collection_name=f"text_collection")
|
119 |
+
index = MultiModalVectorStoreIndex.from_vector_store(vector_store=text_store, image_store=image_store)
|
120 |
+
retriever_engine = index.as_retriever(similarity_top_k=1, image_similarity_top_k=1)
|
121 |
+
shutil.rmtree(temp_dir)
|
122 |
+
return retriever_engine
|
123 |
+
else:
|
124 |
+
temp_dir = tempfile.mkdtemp()
|
125 |
+
temp_pdf_path = os.path.join(temp_dir, pdf_file.name)
|
126 |
+
with open(temp_pdf_path, "wb") as f:
|
127 |
+
f.write(pdf_file.getvalue())
|
128 |
+
|
129 |
+
data_path = os.path.join(temp_dir, "data")
|
130 |
+
os.makedirs(data_path , exist_ok=True)
|
131 |
+
img_save_path = os.path.join(temp_dir, "images")
|
132 |
+
os.makedirs(img_save_path , exist_ok=True)
|
133 |
+
|
134 |
+
extracted_text = extract_text_from_pdf(temp_pdf_path)
|
135 |
+
with open(os.path.join(data_path, "content.txt"), "w") as file:
|
136 |
+
file.write(extracted_text)
|
137 |
+
|
138 |
+
extract_images_from_pdf(temp_pdf_path, img_save_path)
|
139 |
+
moved_count = move_images(img_save_path, data_path)
|
140 |
+
print("Images moved count : ", moved_count)
|
141 |
+
remove_low_size_images(data_path)
|
142 |
+
remove_duplicate_images(data_path)
|
143 |
+
shutil.rmtree(img_save_path)
|
144 |
+
retriever_engine = initialize_qdrant(temp_dir=temp_dir, aws_prefix=aws_prefix_path) # os.path.join("folder" , os.path.splitext(pdf_file.name)[0] , unique_folder_name)
|
145 |
+
upload_folder_to_s3(temp_dir, aws_prefix_path)
|
146 |
+
shutil.rmtree(temp_dir)
|
147 |
+
|
148 |
+
return retriever_engine
|
qdrant_mm_db_pipeline/.lock
DELETED
@@ -1 +0,0 @@
|
|
1 |
-
tmp lock file
|
|
|
|
qdrant_mm_db_pipeline/collection/image_collection_pipeline/storage.sqlite
DELETED
Binary file (307 kB)
|
|
qdrant_mm_db_pipeline/collection/text_collection_pipeline/storage.sqlite
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:abc3d5e877a46fab7aef3c35b28f465d363af0f37e6a7097a4f25b578d941ff7
|
3 |
-
size 3084288
|
|
|
|
|
|
|
|
qdrant_mm_db_pipeline/meta.json
DELETED
@@ -1 +0,0 @@
|
|
1 |
-
{"collections": {"text_collection_pipeline": {"vectors": {"size": 1536, "distance": "Cosine", "hnsw_config": null, "quantization_config": null, "on_disk": null, "datatype": null, "multivector_config": null}, "shard_number": null, "sharding_method": null, "replication_factor": null, "write_consistency_factor": null, "on_disk_payload": null, "hnsw_config": null, "wal_config": null, "optimizers_config": null, "init_from": null, "quantization_config": null, "sparse_vectors": null}, "image_collection_pipeline": {"vectors": {"size": 512, "distance": "Cosine", "hnsw_config": null, "quantization_config": null, "on_disk": null, "datatype": null, "multivector_config": null}, "shard_number": null, "sharding_method": null, "replication_factor": null, "write_consistency_factor": null, "on_disk_payload": null, "hnsw_config": null, "wal_config": null, "optimizers_config": null, "init_from": null, "quantization_config": null, "sparse_vectors": null}}, "aliases": {}}
|
|
|
|
qdrant_mm_db_pipeline_tmp0uzyg0nb_construction_pdf/.lock
DELETED
@@ -1 +0,0 @@
|
|
1 |
-
tmp lock file
|
|
|
|
qdrant_mm_db_pipeline_tmp0uzyg0nb_construction_pdf/collection/image_collection_pipeline_tmp0uzyg0nb_construction_pdf/storage.sqlite
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:716020e47d41d39ea7e3e302ac0687b1a62e2b94991bc700fb5bcbaf6722380f
|
3 |
-
size 1585152
|
|
|
|
|
|
|
|
qdrant_mm_db_pipeline_tmp0uzyg0nb_construction_pdf/collection/text_collection_pipeline_tmp0uzyg0nb_construction_pdf/storage.sqlite
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:c077e65278dfde8d472135f12860bf39fedcf018d44a00d54fc04b2f4eb6afb0
|
3 |
-
size 2998272
|
|
|
|
|
|
|
|
qdrant_mm_db_pipeline_tmp0uzyg0nb_construction_pdf/meta.json
DELETED
@@ -1 +0,0 @@
|
|
1 |
-
{"collections": {"text_collection_pipeline_tmp0uzyg0nb_construction_pdf": {"vectors": {"size": 1536, "distance": "Cosine", "hnsw_config": null, "quantization_config": null, "on_disk": null, "datatype": null, "multivector_config": null}, "shard_number": null, "sharding_method": null, "replication_factor": null, "write_consistency_factor": null, "on_disk_payload": null, "hnsw_config": null, "wal_config": null, "optimizers_config": null, "init_from": null, "quantization_config": null, "sparse_vectors": null}, "image_collection_pipeline_tmp0uzyg0nb_construction_pdf": {"vectors": {"size": 512, "distance": "Cosine", "hnsw_config": null, "quantization_config": null, "on_disk": null, "datatype": null, "multivector_config": null}, "shard_number": null, "sharding_method": null, "replication_factor": null, "write_consistency_factor": null, "on_disk_payload": null, "hnsw_config": null, "wal_config": null, "optimizers_config": null, "init_from": null, "quantization_config": null, "sparse_vectors": null}}, "aliases": {}}
|
|
|
|
requirements.txt
CHANGED
@@ -12,4 +12,5 @@ pillow==10.4.0
|
|
12 |
imagehash
|
13 |
llama-index-embeddings-clip
|
14 |
git+https://github.com/openai/CLIP.git
|
15 |
-
python-dotenv
|
|
|
|
12 |
imagehash
|
13 |
llama-index-embeddings-clip
|
14 |
git+https://github.com/openai/CLIP.git
|
15 |
+
python-dotenv
|
16 |
+
boto3
|
retrieve_and_display.py
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from llama_index.llms.openai import OpenAI
|
2 |
+
from llama_index.core import load_index_from_storage, get_response_synthesizer
|
3 |
+
import matplotlib.pyplot as plt
|
4 |
+
import os
|
5 |
+
from PIL import Image
|
6 |
+
from llama_index.core import PromptTemplate
|
7 |
+
from awsfunctions import download_files_from_s3, check_file_exists_in_s3
|
8 |
+
import tempfile, shutil
|
9 |
+
import streamlit as st
|
10 |
+
|
11 |
+
st.cache_resource()
|
12 |
+
def get_image_from_s3(image_path):
|
13 |
+
temp_dir = tempfile.mkdtemp()
|
14 |
+
download_files_from_s3(temp_dir, [image_path])
|
15 |
+
image = Image.open(os.path.join(temp_dir, image_path))
|
16 |
+
shutil.rmtree(temp_dir)
|
17 |
+
return image
|
18 |
+
|
19 |
+
def plot_images(image_paths):
|
20 |
+
images_shown = 0
|
21 |
+
plt.figure(figsize=(16, 9))
|
22 |
+
for img_path in image_paths:
|
23 |
+
if check_file_exists_in_s3(img_path):
|
24 |
+
image = get_image_from_s3(img_path)
|
25 |
+
plt.subplot(2, 3, images_shown + 1)
|
26 |
+
plt.imshow(image)
|
27 |
+
plt.xticks([])
|
28 |
+
plt.yticks([])
|
29 |
+
images_shown += 1
|
30 |
+
if images_shown >= 6:
|
31 |
+
break
|
32 |
+
|
33 |
+
def retrieve_and_query(query, retriever_engine):
|
34 |
+
retrieval_results = retriever_engine.retrieve(query)
|
35 |
+
|
36 |
+
qa_tmpl_str = (
|
37 |
+
"Context information is below.\n"
|
38 |
+
"---------------------\n"
|
39 |
+
"{context_str}\n"
|
40 |
+
"---------------------\n"
|
41 |
+
"Given the context information , "
|
42 |
+
"answer the query in detail.\n"
|
43 |
+
"Query: {query_str}\n"
|
44 |
+
"Answer: "
|
45 |
+
)
|
46 |
+
qa_tmpl = PromptTemplate(qa_tmpl_str)
|
47 |
+
|
48 |
+
llm = OpenAI(model="gpt-4o", temperature=0)
|
49 |
+
response_synthesizer = get_response_synthesizer(response_mode="refine", text_qa_template=qa_tmpl, llm=llm)
|
50 |
+
|
51 |
+
response = response_synthesizer.synthesize(query, nodes=retrieval_results)
|
52 |
+
|
53 |
+
retrieved_image_path_list = []
|
54 |
+
for node in retrieval_results:
|
55 |
+
if (node.metadata['file_type'] == 'image/jpeg') or (node.metadata['file_type'] == 'image/png'):
|
56 |
+
if node.score > 0.25:
|
57 |
+
retrieved_image_path_list.append(node.metadata['file_path'])
|
58 |
+
|
59 |
+
return response, retrieved_image_path_list
|