Spaces:
Sleeping
Sleeping
File size: 18,162 Bytes
18586e1 505aafa 18586e1 7ff2d8b 18586e1 927dc9c 18586e1 a8087b3 18586e1 7ff2d8b 927dc9c a8087b3 7ff2d8b a8087b3 7ff2d8b 03892d8 7ff2d8b 18586e1 7ff2d8b 18586e1 7ff2d8b 18586e1 7ff2d8b 18586e1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 |
import streamlit as st
import base64
import os
import json
from dotenv import load_dotenv
from openai import OpenAI
#langchain imports
from langchain.chains import RetrievalQA
from langchain.text_splitter import CharacterTextSplitter
from langchain_community.document_loaders import TextLoader
from langchain_community.vectorstores import Chroma
from langchain_openai import OpenAIEmbeddings
from langchain.chains import create_qa_with_sources_chain
from langchain.chains.combine_documents.stuff import StuffDocumentsChain
from langchain.prompts import PromptTemplate
from langchain_openai import ChatOpenAI
from langchain.schema import Document
from langchain.chains.query_constructor.base import AttributeInfo
from langchain.retrievers.self_query.base import SelfQueryRetriever
#from pdf2image import convert_from_path
import io
import json
import re
# Function to encode the image to base64
# def encode_image(image_file):
# return base64.b64encode(image_file.getvalue()).decode("utf-8")
def encode_image(image_files):
base64_images = []
for image_file in image_files:
base64_images.append(base64.b64encode(image_file.getvalue()).decode("utf-8"))
return base64_images, image_files
st.set_page_config(page_title="Document/Image AI Analyst", layout="centered", initial_sidebar_state="collapsed")
# Streamlit page setup
st.title("Document/Image AI Analyst: `GPT-4 with Vision v2` π")
st.write("")
st.write("")
st.write("")
# Guide
st.subheader('What can it do?')
with st.expander('Read more details', expanded=False):
#st.write("There are various use cases that the AI analyst can do!")
st.markdown('- It can describe details found on the images. For instance, it can tell the details on an uploaded passport, such as full name, passport number, country, etc.')
st.markdown('- It can check for validity of images or identification documents. It also checks the legitimacy of documents (if applicable). `Try uploading a suspicious passport picture!`')
st.markdown("- It can compare multiple documents, such as identifying whether a person's photo is the same as the uploaded personal documents. In a comparison use case, feel free to provide extra info (optional) on what comparison you want to perform.")
st.markdown('- And anything else! For a simpler, general demo, upload any image and let it describe what it sees!')
if st.button('Happy prompting and Cheers! π'):
st.balloons()
# new line space
st.write("")
st.subheader('1. We need an OpenAI API key: ')
api_key = st.text_input('',placeholder='Enter your OpenAI API key', type='password', help="You can find your OpenAI API key here: https://platform.openai.com/api-keys. Or if you are provided with one by your organization.")
st.caption('Never share your OpenAI API key to anyone. Note that usage of your OpenAI API key will be billed to your OpenAI account. Keep in mind that an image analysis costs approximately `$0.04`')
# Initialize messages object
messages = []
# Retrieve the OpenAI API Key from secrets
# load_dotenv()
# api_key = os.getenv("OPENAI_API_KEY")
# new line space
st.write("")
# File uploader allows user to add their own image
st.subheader('2. Upload Images: ')
uploaded_files = st.file_uploader("", help='Up to five images only.', type=["jpg", "png", "jpeg"], accept_multiple_files=True)
if api_key:
# Initialize the OpenAI client with the API key
client = OpenAI(api_key=api_key)
# load from disk
docsearch = Chroma(persist_directory="./knowledge_base", embedding_function=OpenAIEmbeddings(api_key=api_key))
metadata_field_info = [
AttributeInfo(
name="type",
description="The type of document",
type="string",
),
AttributeInfo(
name="name",
description="The name of the document",
type="string",
),
AttributeInfo(
name="filename",
description="The file name of the document",
type="string",
),
]
document_content_description = "Type of a document"
llm = ChatOpenAI(temperature=0, api_key=api_key)
retriever = SelfQueryRetriever.from_llm(
llm,
docsearch,
document_content_description,
metadata_field_info,
enable_limit=True,
search_kwargs={"k": 1}
)
# This sample for querying
# results = retriever.invoke("requirements in passport")
# st.write("RESULTS: ", results)
#############
if uploaded_files:
if len(uploaded_files) <= 5:
st.success("You uploaded " + str(len(uploaded_files)) + " images!", icon="β
")
elif len(uploaded_files) > 5:
st.error("More than 5 uploaded images. Please remove.", icon="β")
file_names = [file.name for file in uploaded_files]
print("Filenames: ", file_names)
for uploaded_file in uploaded_files:
with st.expander("Uploaded image: `" + uploaded_file.name + "`", expanded = False):
st.image(uploaded_file, use_column_width=True)
# if uploaded_file:
# # Display the uploaded image
# with st.expander("Image", expanded = True):
# st.image(uploaded_file, caption=uploaded_file.name, use_column_width=True)
# new line space
st.write("")
# Toggle for showing additional details input
st.subheader('3. Details about the images:')
show_details = st.toggle("Add details about the images (optional)", value=False)
st.caption('')
if show_details:
# Text input for additional details about the image, shown only if toggle is True
additional_details = st.text_area(
"Add any additional details or context about the image(s) here:",
placeholder='I am typically able to understand images without context, but feel free to describe what type of analysis you want. For instance, verifying personal documents, checking for falsification or nothing at all (optional)',
disabled=not show_details
)
# new line space
st.write("")
# Button to trigger the analysis
st.subheader('4. Analyze! ')
analyze_button = st.button("Analyse the image(s)", type="secondary")
st.caption('')
# Check if an image has been uploaded, if the API key is available, and if the button has been pressed
if uploaded_files is not None and api_key and analyze_button:
with st.spinner("Analysing the image(s) ..."):
# Encode the image
base64_image, filenames = encode_image(uploaded_files)
#print("base64 images: ", filenames)
#kb_details = ("")
# Optimized prompt for additional clarity and detail
prompt_text = (
"You are a highly knowledgeable student admission document identifier. Evaluate each document and identify what type of admission document it is. "
"Write your output as a JSON object mapping of the document file name and the type of document you identified, using double-quote, not single-quote. "
"The format of the output must be enclosed in curly-brackets. "
f"\n\nYou are provided with these corresponding document file names in order:\n{file_names}"
)
if show_details and additional_details:
prompt_text += (
f"\n\nAdditional Context Provided by the User:\n{additional_details}"
#f"\n\nGeneral Admissions Guide on how to verify student admission documents:\n{kb_details}"
)
# IF scenarios for images payload for messages var
if len(uploaded_files) == 1:
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": prompt_text},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[0]}",
},
],
}
]
elif len(uploaded_files) == 2:
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": prompt_text},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[0]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[1]}",
},
],
}
]
elif len(uploaded_files) == 3:
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": prompt_text},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[0]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[1]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[2]}",
},
],
}
]
elif len(uploaded_files) == 4:
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": prompt_text},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[0]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[1]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[2]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[3]}",
},
],
}
]
elif len(uploaded_files) == 5:
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": prompt_text},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[0]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[1]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[2]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[3]}",
},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[4]}",
},
],
}
]
elif len(uploaded_files) > 5:
messages = []
# Make the request to the OpenAI API
try:
##################
# STEP 1: Identify and map filenames to type of document
full_response = ''
message_placeholder = st.empty()
for completion in client.chat.completions.create(
model="gpt-4-vision-preview", messages=messages,
max_tokens=1200, stream=True,
):
# Check if there is content to display
if completion.choices[0].delta.content is not None:
full_response += completion.choices[0].delta.content
message_placeholder.markdown(full_response + "β")
# Final update to placeholder after the stream ends
#message_placeholder.text(full_response)
#full_response = '{"filename1": "Give full information about IELTS Test Result without summarizing", "filename2": "Give full information about TOEFL Test results without summarizing"}'
##################
# STEP 2: perform validation via KB retrieval
# STEP 2.1: Retrieval individual info and concat into an array
# Clean up format:
json_response = re.search(r'\{.*\}', full_response, re.DOTALL).group()
full_response_json = json.loads(json_response)
# Display identified documents
# Initialize an empty list to store markdown lines
markdown_list = []
# Iterate through the dictionary and format each item
for filename, content in full_response_json.items():
markdown_list.append(f"- {filename} is a {content}")
# Join the list into a single string with newline characters
markdown_string = '\n'.join(markdown_list)
message_placeholder.markdown("**Identified Documents:**\n\n" + markdown_string)
### TO DO : The ABOVE code can be in a function or object.
# Initialize an empty array to store the full prompt as log
retrieved_knowledge = []
# Iterate over each item in the full_response dictionary
st.subheader("5. Analysis Results")
for i, (filename, details) in enumerate(full_response_json.items()):
with st.expander(details):
file_content = ""
# perform query to KB, details is the query
results = retriever.invoke(details)
# Step 1: Access the first Document
document = results[0]
# Step 2: Retrieve the filename from the metadata
kb_retrieved_filename = document.metadata['filename']
# Step 3: Open the file using the extracted filename
with open(kb_retrieved_filename, 'r') as file:
file_content = file.read()
### TO DO: NOW perform the actual validation
prompt_text_validation = (
"You are a highly knowledgeable student admission document image verifier. "
"Your task is to examine the image in detail. "
#"Provide a comprehensive, factual, and accurate explanation of what each image depict and classify them based on which student admission document it is. "
"Verify and compare whether the details in the document image is valid by checking the provided DOCUMENT REQUIREMENTS. "
"Highlight key elements on the document image that is valid and invalid, and present your analysis in clear, well-structured markdown format. "
"If applicable, identify any falsification, tampering and editing of the image that could potentially mean the document is not valid and was tampered. "
#"Assume the reader has a basic understanding of how student admission documents should be."
"Lastly, include your final verdict on whether the document is legit or needs further checking. Label as [LEGIT], [NEED FURTHER CHECKING] or [NOT LEGIT]"
f"\n\nDOCUMENT REQUIREMENTS that must be met to be considered valid:\n{file_content}"
)
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": prompt_text_validation},
{
"type": "image_url",
"image_url": f"data:image/jpeg;base64,{base64_image[i]}",
},
],
}
]
validation_response = ''
validation_placeholder = st.empty()
prompt_message = ''
#prompt_placeholder = st.write("COMPLETE PROMPT", prompt_text_validation)
for completion in client.chat.completions.create(
model="gpt-4-vision-preview", messages=messages,
max_tokens=1200, stream=True
):
# Check if there is content to display
if completion.choices[0].delta.content is not None:
validation_response += completion.choices[0].delta.content
validation_placeholder.markdown(validation_response + "β")
# Final update to validation placeholder after the stream ends
#validation_placeholder.text(validation_response)
# Now you can work with file_content
#print(file_content)
# add data to array
retrieved_knowledge.extend([prompt_text_validation, messages])
validation_placeholder.markdown(validation_response)
with st.expander("Process Logs: "):
st.json(retrieved_knowledge)
# Display the response in the app
# st.write(response.choices[0].message.content)
except Exception as e:
st.error(f"An error occurred: {e}")
else:
# Warnings for user action required
if not uploaded_files and analyze_button:
st.warning("Please upload at least one image. Up to five.", icon="β οΈ")
if not api_key:
st.error("Please enter your OpenAI API key.", icon="β") |