File size: 14,168 Bytes
f4a3eb8
 
 
 
 
44c1a2a
f4a3eb8
 
bfe51e4
f4a3eb8
e789621
44c1a2a
bfe51e4
f4a3eb8
e789621
f4a3eb8
 
e56b39d
 
e789621
f4a3eb8
e789621
 
f4a3eb8
 
 
e56b39d
bfe51e4
 
 
 
 
e56b39d
bfe51e4
e56b39d
44c1a2a
e789621
 
e56b39d
e789621
f4a3eb8
e56b39d
bfe51e4
e56b39d
 
44c1a2a
bfe51e4
 
 
 
 
 
 
 
 
44c1a2a
 
e56b39d
 
bfe51e4
e56b39d
 
44c1a2a
 
e56b39d
 
bfe51e4
e56b39d
 
44c1a2a
e56b39d
44c1a2a
 
bfe51e4
 
 
 
 
 
 
 
 
 
44c1a2a
 
bfe51e4
 
 
44c1a2a
 
 
 
 
bfe51e4
 
e56b39d
 
 
bfe51e4
e56b39d
44c1a2a
e56b39d
e789621
bfe51e4
e789621
 
 
 
 
 
bfe51e4
 
 
 
 
e789621
f4a3eb8
e789621
 
 
 
f4a3eb8
e789621
 
44c1a2a
f4a3eb8
e789621
f4a3eb8
 
e789621
44c1a2a
 
 
f4a3eb8
 
e56b39d
bfe51e4
e56b39d
f4a3eb8
 
bfe51e4
f4a3eb8
e789621
 
e56b39d
 
 
f4a3eb8
e56b39d
 
e789621
f4a3eb8
e789621
bfe51e4
e789621
 
 
 
 
 
44c1a2a
f4a3eb8
e56b39d
bfe51e4
 
 
e56b39d
bfe51e4
 
 
e789621
 
 
 
 
bfe51e4
 
 
 
e789621
 
 
 
 
bfe51e4
 
 
 
e789621
e56b39d
 
bfe51e4
 
 
 
 
 
 
f4a3eb8
44c1a2a
e56b39d
e789621
 
e56b39d
f4a3eb8
e789621
f4a3eb8
e789621
f4a3eb8
bfe51e4
44c1a2a
bfe51e4
 
 
44c1a2a
 
bfe51e4
 
44c1a2a
 
 
bfe51e4
e56b39d
44c1a2a
f4a3eb8
e789621
bfe51e4
e789621
f4a3eb8
 
e789621
 
f4a3eb8
e789621
f4a3eb8
e789621
 
 
f4a3eb8
bfe51e4
 
e56b39d
 
 
 
44c1a2a
 
e56b39d
bfe51e4
44c1a2a
e56b39d
bfe51e4
44c1a2a
e56b39d
 
bfe51e4
 
44c1a2a
bfe51e4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44c1a2a
bfe51e4
 
44c1a2a
 
e56b39d
e789621
bfe51e4
e789621
f4a3eb8
bfe51e4
f4a3eb8
44c1a2a
f4a3eb8
 
e789621
bfe51e4
f4a3eb8
 
e56b39d
f4a3eb8
 
 
 
e56b39d
f4a3eb8
e56b39d
f4a3eb8
 
 
 
bfe51e4
 
 
 
f4a3eb8
 
 
e56b39d
bfe51e4
 
 
f4a3eb8
e789621
bfe51e4
f4a3eb8
bfe51e4
 
e56b39d
f4a3eb8
e789621
bfe51e4
f4a3eb8
bfe51e4
 
e789621
f4a3eb8
bfe51e4
 
 
 
e789621
 
 
f4a3eb8
bfe51e4
f4a3eb8
bfe51e4
f4a3eb8
bfe51e4
 
 
f4a3eb8
bfe51e4
 
 
 
 
 
 
e56b39d
bfe51e4
e56b39d
bfe51e4
f4a3eb8
e789621
bfe51e4
 
 
 
 
e56b39d
bfe51e4
 
e56b39d
f4a3eb8
 
e789621
 
44c1a2a
f4a3eb8
44c1a2a
f4a3eb8
bfe51e4
 
 
 
 
 
e789621
 
e56b39d
bfe51e4
 
e56b39d
 
bfe51e4
 
 
 
 
e56b39d
bfe51e4
f4a3eb8
bfe51e4
 
 
 
 
 
f4a3eb8
bfe51e4
f4a3eb8
 
bfe51e4
 
 
 
f4a3eb8
e789621
f4a3eb8
 
bfe51e4
 
 
 
 
 
e56b39d
bfe51e4
 
 
 
 
 
 
 
 
 
 
f4a3eb8
 
e789621
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
434
435
436
437
438
439
440
441
442
443
444
import gradio as gr
from pathlib import Path
from PIL import Image
import PyPDF2
import docx
from sentence_transformers import SentenceTransformer, util
import faiss
import numpy as np
from transformers import AutoTokenizer, AutoModelForCausalLM, BlipProcessor, BlipForConditionalGeneration
import torch
from datetime import datetime
import fitz  # PyMuPDF
import shutil

# Load models
print("Loading models...")
embedding_model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')

print("Loading LLM...")
tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v1.0")
llm_model = AutoModelForCausalLM.from_pretrained(
    "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
    torch_dtype=torch.float16,
    device_map="auto"
)

print("Loading image caption model...")
caption_processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
caption_model = BlipForConditionalGeneration.from_pretrained(
    "Salesforce/blip-image-captioning-large",
    torch_dtype=torch.float16
).to("cuda" if torch.cuda.is_available() else "cpu")

print("βœ… All models loaded!")

# Storage
documents = []
images = []
image_captions = []
embeddings_index = None

def generate_image_caption(image_path):
    """Generate detailed caption for image"""
    try:
        img = Image.open(image_path).convert('RGB')
        
        # Generate detailed caption
        inputs = caption_processor(img, return_tensors="pt").to(caption_model.device)
        output = caption_model.generate(
            **inputs, 
            max_length=100,
            num_beams=5,
            temperature=0.7
        )
        caption = caption_processor.decode(output[0], skip_special_tokens=True)
        
        return caption.strip()
    except Exception as e:
        print(f"Caption error: {e}")
        return ""

def extract_images_from_pdf(pdf_path):
    """Extract images from PDF"""
    extracted = []
    try:
        doc = fitz.open(pdf_path)
        
        for page_num in range(len(doc)):
            page = doc[page_num]
            images_list = page.get_images(full=True)
            
            for img_index, img in enumerate(images_list):
                try:
                    xref = img[0]
                    base_image = doc.extract_image(xref)
                    image_bytes = base_image["image"]
                    
                    # Save image
                    img_path = f"/tmp/pdf_page{page_num+1}_img{img_index}.png"
                    with open(img_path, "wb") as f:
                        f.write(image_bytes)
                    
                    # Validate image
                    test_img = Image.open(img_path)
                    width, height = test_img.size
                    
                    # Only keep meaningful images (not tiny icons/logos)
                    if width >= 150 and height >= 150:
                        extracted.append({
                            'path': img_path,
                            'page': page_num + 1,
                            'source': Path(pdf_path).name
                        })
                except Exception as e:
                    continue
        
        doc.close()
    except Exception as e:
        print(f"PDF image extraction error: {e}")
    
    return extracted

def extract_pdf_text(pdf_path):
    """Extract text from PDF"""
    chunks = []
    with open(pdf_path, 'rb') as f:
        pdf = PyPDF2.PdfReader(f)
        for i, page in enumerate(pdf.pages):
            text = page.extract_text()
            if text.strip():
                chunks.append({
                    'text': text,
                    'page': i + 1,
                    'source': Path(pdf_path).name
                })
    return chunks

def extract_docx_text(docx_path):
    doc = docx.Document(docx_path)
    text = '\n'.join([p.text for p in doc.paragraphs if p.text.strip()])
    return [{'text': text, 'source': Path(docx_path).name}]

def extract_txt_text(txt_path):
    with open(txt_path, 'r', encoding='utf-8') as f:
        return [{'text': f.read(), 'source': Path(txt_path).name}]

def chunk_text(text, size=400):
    words = text.split()
    chunks = []
    for i in range(0, len(words), size):
        chunk = ' '.join(words[i:i+size])
        if chunk.strip():
            chunks.append(chunk)
    return chunks

def process_files(files, progress=gr.Progress()):
    """Process uploaded files"""
    global documents, images, image_captions, embeddings_index
    
    if not files:
        return "⚠️ Please upload files first"
    
    documents = []
    images = []
    image_captions = []
    
    total = len(files)
    
    for idx, file in enumerate(files):
        progress((idx + 1) / total, desc=f"Processing {Path(file.name).name}...")
        ext = Path(file.name).suffix.lower()
        
        if ext == '.pdf':
            # Extract text
            chunks = extract_pdf_text(file.name)
            for chunk in chunks:
                for small_chunk in chunk_text(chunk['text']):
                    documents.append({
                        'text': small_chunk,
                        'source': chunk['source'],
                        'page': chunk['page']
                    })
            
            # Extract images
            pdf_images = extract_images_from_pdf(file.name)
            for img in pdf_images:
                caption = generate_image_caption(img['path'])
                if caption:  # Only add if caption generated
                    images.append(img)
                    image_captions.append(caption)
        
        elif ext == '.docx':
            chunks = extract_docx_text(file.name)
            for chunk in chunks:
                for small_chunk in chunk_text(chunk['text']):
                    documents.append({
                        'text': small_chunk,
                        'source': chunk['source']
                    })
        
        elif ext == '.txt':
            chunks = extract_txt_text(file.name)
            for chunk in chunks:
                for small_chunk in chunk_text(chunk['text']):
                    documents.append({
                        'text': small_chunk,
                        'source': chunk['source']
                    })
        
        elif ext in ['.jpg', '.jpeg', '.png', '.gif', '.bmp']:
            caption = generate_image_caption(file.name)
            if caption:
                images.append({
                    'path': file.name,
                    'source': Path(file.name).name,
                    'page': ''
                })
                image_captions.append(caption)
    
    # Create embeddings
    progress(0.9, desc="Creating embeddings...")
    if documents:
        texts = [doc['text'] for doc in documents]
        embeddings = embedding_model.encode(texts, show_progress_bar=False)
        
        index = faiss.IndexFlatL2(embeddings.shape[1])
        index.add(embeddings.astype('float32'))
        embeddings_index = index
    
    progress(1.0, desc="Done!")
    
    status = f"βœ… **Processing Complete!**\n\n"
    status += f"πŸ“„ **Text chunks:** {len(documents)}\n"
    status += f"πŸ–ΌοΈ **Images extracted:** {len(images)}\n"
    
    if images:
        status += f"\n**Sample captions:**\n"
        for i, (img, cap) in enumerate(zip(images[:3], image_captions[:3]), 1):
            status += f"{i}. {img['source']}"
            if img.get('page'):
                status += f" (Page {img['page']})"
            status += f":\n   _{cap}_\n"
    
    return status

def search_documents(query, k=3):
    """Search relevant documents"""
    if not documents or embeddings_index is None:
        return []
    
    query_vec = embedding_model.encode([query])
    distances, indices = embeddings_index.search(query_vec.astype('float32'), k)
    
    results = []
    for idx in indices[0]:
        if idx < len(documents):
            results.append(documents[idx])
    return results

def find_relevant_images(query, relevance_threshold=0.25):
    """Find images ONLY if relevant to query"""
    if not images or not image_captions:
        return [], []
    
    # Encode query and captions
    query_emb = embedding_model.encode(query, convert_to_tensor=True)
    caption_embs = embedding_model.encode(image_captions, convert_to_tensor=True)
    
    # Calculate cosine similarity
    similarities = util.cos_sim(query_emb, caption_embs)[0]
    
    # Filter by threshold and get top 3
    relevant_imgs = []
    explanations = []
    
    for idx, sim_score in enumerate(similarities):
        sim_value = float(sim_score)
        
        # Only show if relevance > threshold
        if sim_value > relevance_threshold:
            img_info = images[idx]
            caption = image_captions[idx]
            
            relevant_imgs.append(img_info['path'])
            
            # Create explanation
            exp = f"**πŸ“„ Source:** {img_info['source']}"
            if img_info.get('page'):
                exp += f" (Page {img_info['page']})"
            exp += f"\n**πŸ’¬ Description:** {caption}"
            exp += f"\n**🎯 Relevance:** {sim_value * 100:.1f}%\n"
            
            explanations.append(exp)
    
    # Sort by relevance and take top 3
    if relevant_imgs:
        sorted_pairs = sorted(
            zip(similarities, relevant_imgs, explanations),
            key=lambda x: x[0],
            reverse=True
        )[:3]
        
        relevant_imgs = [pair[1] for pair in sorted_pairs]
        explanations = [pair[2] for pair in sorted_pairs]
    
    return relevant_imgs, explanations

def generate_answer(question, context_docs):
    """Generate answer from context"""
    context = '\n\n'.join([doc['text'] for doc in context_docs])
    
    prompt = f"""Answer this question based only on the context provided. Be concise and accurate.

Context:
{context}

Question: {question}

Answer:"""
    
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1200)
    
    with torch.no_grad():
        outputs = llm_model.generate(
            inputs.input_ids,
            max_new_tokens=200,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
    
    answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    # Extract answer part
    if "Answer:" in answer:
        answer = answer.split("Answer:")[-1].strip()
    
    return answer

def answer_query(question, progress=gr.Progress()):
    """Answer question with relevant images only"""
    if not question.strip():
        return "⚠️ Please enter a question", None
    
    if not documents:
        return "⚠️ Please upload and process documents first", None
    
    # Search documents
    progress(0.3, desc="Searching documents...")
    relevant_docs = search_documents(question, k=3)
    
    if not relevant_docs:
        return "❌ No relevant information found", None
    
    # Generate answer
    progress(0.6, desc="Generating answer...")
    answer = generate_answer(question, relevant_docs)
    
    # Format response
    response = f"## πŸ’‘ Answer\n\n{answer}\n\n"
    response += f"## πŸ“š Text Sources\n\n"
    
    for i, doc in enumerate(relevant_docs, 1):
        source = doc['source']
        page = doc.get('page', '')
        if page:
            response += f"{i}. **{source}** (Page {page})\n"
        else:
            response += f"{i}. **{source}**\n"
    
    # Find relevant images
    progress(0.9, desc="Finding relevant images...")
    relevant_imgs, img_explanations = find_relevant_images(question, relevance_threshold=0.25)
    
    # Add image explanations if found
    if relevant_imgs and img_explanations:
        response += f"\n## πŸ–ΌοΈ Related Images\n\n"
        for exp in img_explanations:
            response += f"{exp}\n"
    else:
        response += f"\n_No relevant images found for this query_\n"
    
    progress(1.0, desc="Done!")
    
    return response, relevant_imgs if relevant_imgs else None

# UI
with gr.Blocks(
    title="DocVision AI",
    theme=gr.themes.Soft(primary_hue="indigo")
) as app:
    
    gr.Markdown("""
    # πŸ“š DocVision AI - Intelligent Document Q&A
    ### Upload documents and get AI-powered answers with relevant images
    """)
    
    with gr.Row():
        with gr.Column():
            file_input = gr.File(
                label="πŸ“ Upload Documents & Images",
                file_count="multiple",
                file_types=[".pdf", ".docx", ".txt", ".jpg", ".png"]
            )
            process_btn = gr.Button(
                "⚑ Process Documents",
                variant="primary",
                size="lg"
            )
            status = gr.Markdown(label="πŸ“Š Processing Status")
        
        with gr.Column():
            question = gr.Textbox(
                label="❓ Ask Your Question",
                placeholder="What would you like to know about your documents?",
                lines=3
            )
            ask_btn = gr.Button(
                "πŸ” Get Answer",
                variant="primary",
                size="lg"
            )
    
    answer = gr.Markdown(label="πŸ“ Answer with Sources")
    
    gallery = gr.Gallery(
        label="πŸ–ΌοΈ Relevant Images (Only shown if related to your question)",
        columns=2,
        height=500,
        show_label=True
    )
    
    gr.Markdown("### πŸ’‘ Example Questions")
    gr.Examples(
        examples=[
            ["What is the main topic of this document?"],
            ["Explain the workflow or architecture shown"],
            ["What are the key findings?"],
            ["Describe any diagrams or charts present"]
        ],
        inputs=question
    )
    
    # Event handlers
    process_btn.click(
        process_files,
        inputs=[file_input],
        outputs=[status]
    )
    
    ask_btn.click(
        answer_query,
        inputs=[question],
        outputs=[answer, gallery]
    )
    
    question.submit(
        answer_query,
        inputs=[question],
        outputs=[answer, gallery]
    )

if __name__ == "__main__":
    app.launch()