File size: 27,387 Bytes
161aac1 e17f518 161aac1 e17f518 161aac1 e17f518 161aac1 e17f518 161aac1 e17f518 161aac1 e17f518 161aac1 e17f518 161aac1 e17f518 161aac1 e17f518 161aac1 e17f518 161aac1 24bb722 cdaaa6f e17f518 161aac1 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 |
import gradio as gr
import lancedb
import os
from video_pipeline import run_pipeline
from langchain_openai import ChatOpenAI
from langchain_core.tools import tool
import base64
import tempfile
import shutil
from utils import get_text_embedding, sample_from_video, convert_image_to_base64
from config import load_config
from lancedb_utils import retreive_clip
from gradio import ChatMessage
app_config = load_config()
langchain_message_history = []
chat_model = ChatOpenAI(
# model="Qwen/Qwen3-30B-A3B",
model="Qwen/Qwen3-32B",
base_url="https://api.studio.nebius.com/v1/",
api_key=app_config.NEBIUS_API_KEY.get_secret_value()
)
chat_model_vlm = ChatOpenAI(
model="Qwen/Qwen2.5-VL-32B-Instruct",
base_url=app_config.MODAL_VLM_URL.get_secret_value(),
api_key=app_config.MODEL_API_KEY.get_secret_value()
)
def search_clips(query_text, limit=3):
"""Searches the LanceDB database for clips matching the query."""
try:
# Create embedding for the query using Hugging Face API
query_vector = get_text_embedding(query_text, app_config.CLIP_EMBEDDING_URL.get_secret_value())[0]
# Connect to LanceDB
db = lancedb.connect(app_config.LANCEDB_URI.get_secret_value())
table = db.open_table("video_clips")
# Search for similar clips
results = table.search(query_vector).limit(limit).to_pandas()
return results
except FileNotFoundError:
return f"Error: Database not found at {app_config.LANCEDB_URI.get_secret_value()}. Please ensure the video analysis server has processed some videos first."
except Exception as e:
return f"Error during search: {str(e)}"
def format_search_results(results_df):
"""Format search results for display."""
if isinstance(results_df, str): # Error message
return results_df
if results_df.empty:
return "No clips found matching your query."
response = "Here are the top results I found:\n\n"
for idx, row in results_df.iterrows():
response += f"**Clip {row.get('clip_id', 'N/A')} from {row.get('video_name', 'Unknown')}**\n"
response += f"β° Time: {row.get('start_time', 'N/A')}s - {row.get('end_time', 'N/A')}s\n"
# Handle summary safely
summary = row.get('summary', 'No summary available')
if isinstance(summary, str) and '---' in summary:
summary = summary.split('---')[0].strip()
response += f"π Summary: {summary}\n"
# Add score if available
if '_distance' in row:
score = 1 - row['_distance'] # Convert distance to similarity score
response += f"π― Relevance: {score:.2f}\n"
response += "\n---\n\n"
return response
def get_clip_videos_and_thumbnails(results_df):
"""Extract video clips and thumbnails from search results."""
if isinstance(results_df, str) or results_df.empty:
return [], []
videos = []
thumbnails = []
for idx, row in results_df.iterrows():
# Get video clip path
clip_path = row.get('clip_path', '')
if clip_path and os.path.exists(clip_path):
videos.append(clip_path)
else:
videos.append(None)
# Get thumbnail from base64
thumbnail_b64 = row.get('thumbnail', '')
if thumbnail_b64:
try:
# Decode base64 thumbnail and save as temp file
thumbnail_data = base64.b64decode(thumbnail_b64)
temp_thumb = tempfile.NamedTemporaryFile(delete=False, suffix='.jpg')
temp_thumb.write(thumbnail_data)
temp_thumb.close()
thumbnails.append(temp_thumb.name)
except Exception as e:
print(f"Error processing thumbnail: {e}")
thumbnails.append(None)
else:
thumbnails.append(None)
return videos, thumbnails
# Global state to store the latest search results for the UI
latest_search_results = {"results": None, "query": "", "clips_display": []}
@tool
def get_relevant_clips(query):
"""Retrieve relevant clips from vector database
Args:
query: Text to use in vector search
Returns :
str: the search results formatted in a string
"""
global latest_search_results
search_result = search_clips(query, limit=5)
formatted_search_result = format_search_results(search_result)
# Store the results globally so the UI can access them
latest_search_results["results"] = search_result
latest_search_results["query"] = query
# Prepare clips display data
if not isinstance(search_result, str) and not search_result.empty:
videos, thumbnails = get_clip_videos_and_thumbnails(search_result)
clip_components = []
for idx, (row_idx, row) in enumerate(search_result.iterrows()):
video = videos[idx] if idx < len(videos) else None
thumbnail = thumbnails[idx] if idx < len(thumbnails) else None
if video or thumbnail: # Only show if we have media
info = {
'clip_id': row.get('clip_id', 'N/A'),
'video_name': row.get('video_name', 'Unknown'),
'start_time': row.get('start_time', 'N/A'),
'end_time': row.get('end_time', 'N/A'),
'summary': row.get('summary', '').split('---')[0].strip() if '---' in str(row.get('summary', '')) else row.get('summary', ''),
'relevance': 1 - row['_distance'] if '_distance' in row else 0
}
clip_components.append({
'video': video,
'thumbnail': thumbnail,
'info': info
})
latest_search_results["clips_display"] = clip_components
else:
latest_search_results["clips_display"] = []
return formatted_search_result
@tool
def get_clip(clip_id: str):
"""Retreive the clip
Args:
clip_id: id of the clip to retreive
Returns :
list: list of frames
"""
print("clip id", clip_id)
clip = retreive_clip(clip_id, app_config.LANCEDB_URI.get_secret_value())
images = sample_from_video(clip["clip_path"])
base64_images = [convert_image_to_base64(image, "png") for image in images]
return base64_images
def search_and_display_clips(query_text):
"""Search for clips and return both formatted text and video/thumbnail data."""
search_results = search_clips(query_text, limit=5)
formatted_results = format_search_results(search_results)
if isinstance(search_results, str): # Error case
return formatted_results, [], []
videos, thumbnails = get_clip_videos_and_thumbnails(search_results)
# Prepare clip info for display
clip_info = []
for idx, row in search_results.iterrows():
info = {
'clip_id': row.get('clip_id', 'N/A'),
'video_name': row.get('video_name', 'Unknown'),
'start_time': row.get('start_time', 'N/A'),
'end_time': row.get('end_time', 'N/A'),
'summary': row.get('summary', '').split('---')[0].strip() if '---' in str(row.get('summary', '')) else row.get('summary', ''),
'relevance': 1 - row['_distance'] if '_distance' in row else 0
}
clip_info.append(info)
return formatted_results, videos, thumbnails, clip_info
def chat_agent(message, history: list):
"""Core agent logic function."""
global latest_search_results, langchain_message_history
# Add current message
langchain_message_history.append({"role": "user", "content": message})
llm_with_tool = chat_model.bind_tools(tools=[get_relevant_clips])
tools = {"get_relevant_clips": get_relevant_clips}
# The agent loop
while True:
ai_response = llm_with_tool.invoke(langchain_message_history)
if not ai_response.tool_calls:
break
for tool_call in ai_response.tool_calls:
tool_output = tools[tool_call["name"]].invoke(tool_call)
tool_call_log = {
"role": "tool",
"tool_call_id": tool_output.tool_call_id,
"content": tool_output.content
}
langchain_message_history.append(tool_call_log)
content = ai_response.content
if "</think>" in content:
content = content.split("</think>")[-1].strip()
# The global state `latest_search_results` is updated by the tool.
# The text response is returned.
langchain_message_history.append({"role": "assistant", "content": content})
return langchain_message_history
def chat_agent_mm(message, history):
"""Core agent logic function."""
global latest_search_results, langchain_message_history
langchain_message_history.append({"role": "user", "content": message})
history.append({"role": "user", "content": message})
print(langchain_message_history)
llm_with_tool = chat_model_vlm.bind_tools(tools=[get_relevant_clips, get_clip])
tools = {"get_relevant_clips": get_relevant_clips, "get_clip": get_clip}
# The agent loop
while True:
ai_response = llm_with_tool.invoke(langchain_message_history)
if not ai_response.tool_calls:
break
for tool_call in ai_response.tool_calls:
print(tool_call)
langchain_message_history.append(
{
"role": "assistant",
"content": "",
"tool_calls": [
tool_call
]
}
)
history.append(
{
"role": "assistant",
"content": "",
"tool_calls": [
tool_call
]
}
)
tool_output = tools[tool_call["name"]].invoke(tool_call)
if tool_call["name"] == "get_clip":
tool_call_log = {
"role": "tool",
"tool_call_id": tool_output.tool_call_id,
"content": "retrieved clip will be provided by the user after this message"
}
history.append(tool_call_log)
langchain_message_history.extend([
tool_call_log,
{
"role": "user", "content": [
{"type": "text", "text": "here is the clip retreived by the tool"},
*map(lambda x: {"type": "image_url", "image_url": {"url": f'data:image/png;base64,{x}'}}, tool_output.content)
],
}])
else:
tool_call_log = {
"role": "tool",
"tool_call_id": tool_output.tool_call_id,
"content": tool_output.content
}
langchain_message_history.append(tool_call_log)
history.append(tool_call_log)
content = ai_response.content
if "</think>" in content:
content = content.split("</think>")[-1].strip()
# The global state `latest_search_results` is updated by the tool.
# The text response is returned.
langchain_message_history.append({"role": "assistant", "content": content})
history.append({"role": "assistant", "content": content})
return history
def get_latest_clips_for_display():
"""Get the latest search results for display in the UI."""
global latest_search_results
return latest_search_results.get("clips_display", [])
def check_database_status():
"""Check if the database exists and has data."""
try:
db = lancedb.connect(app_config.LANCEDB_URI.get_secret_value())
table_names = db.table_names()
if "video_clips" not in table_names:
return f"β
Database connected, but 'video_clips' table not found. Analyze a video to create it."
table = db.open_table("video_clips")
count = len(table)
return f"β
Database connected. Found {count} video clips."
except Exception as e:
return f"β Database issue: {str(e)}"
def check_server_status():
"""Check if the MCP server is running."""
# This check is illustrative; adjust if your server runs on a different port.
return "β‘οΈ To analyze videos, upload them in the 'Video Analyzer' tab."
# Create the Gradio interface
with gr.Blocks(title="Video Search Agent", theme=gr.themes.Soft()) as demo:
gr.Markdown("# π€ Video Search Agent")
gr.Markdown("Search through your processed video clips using natural language queries.")
# Status section
with gr.Accordion("System Status", open=False):
status_text = gr.Textbox(
label="Status",
value=f"{check_database_status()}\n{check_server_status()}",
interactive=False,
lines=3
)
refresh_btn = gr.Button("Refresh Status")
refresh_btn.click(
fn=lambda: f"{check_database_status()}\n{check_server_status()}",
outputs=status_text
)
# Chat interface with clips display
with gr.Tab("π¬ Chat with Clips"):
# Manual chat layout for full control
chatbot = gr.Chatbot(
[],
type="messages",
label="Video Search Assistant",
height=500,
avatar_images=(None, "https://seeklogo.com/images/O/openai-logo-8284262873-seeklogo.com.png")
)
with gr.Row():
chat_input = gr.Textbox(
show_label=False,
placeholder="Ask me to find clips about cooking...",
lines=1,
scale=4,
container=False,
)
submit_btn = gr.Button("π Search", variant="primary", scale=1, min_width=150)
gr.Examples(
[
"find clips about cooking",
"search for meeting discussions",
"show me sports highlights",
"find outdoor activities"
],
inputs=chat_input,
label="Quick-search examples"
)
gr.Markdown("### π¬ Found Clips")
# State to store clips data for rendering
clips_data_state = gr.State([])
def handle_chat_and_clips(user_message, history):
"""Event handler for chat submission to update chat and clips."""
new_history = chat_agent(user_message, history)
# print(new_history)
clips_data = get_latest_clips_for_display()
return "", new_history, clips_data
# Dynamic clip display
@gr.render(inputs=[clips_data_state])
def show_clips_in_chat(clip_data):
# FIX: Wrap everything in a single gr.Column to ensure vertical stacking
with gr.Column():
if not clip_data:
gr.Markdown("*No clips found yet. Ask the assistant to search for something!*")
return
gr.Markdown(f"**Found {len(clip_data)} relevant clips:**")
for i, clip in enumerate(clip_data):
# Use a column for each clip block to keep them separate
with gr.Column(variant='panel'):
with gr.Row():
with gr.Column(scale=3):
# Clip info
info = clip['info']
gr.Markdown(f"**Clip {info['clip_id']}** from *{info['video_name']}*")
gr.Markdown(f"β±οΈ {info['start_time']:.1f}s - {info['end_time']:.1f}s | π― Relevance: {info['relevance']:.2f}")
# Summary (shortened for chat view)
summary_text = info['summary'][:150] + "..." if len(info['summary']) > 150 else info['summary']
gr.Markdown(f"π {summary_text}")
with gr.Column(scale=1):
# Video player (smaller for chat view)
if clip['video'] and os.path.exists(clip['video']):
gr.Video(clip['video'], label="", height=180, show_label=False)
else:
gr.Markdown("β οΈ *Video not available*")
# Wire up submission events
submit_btn.click(
fn=handle_chat_and_clips,
inputs=[chat_input, chatbot],
outputs=[chat_input, chatbot, clips_data_state]
)
chat_input.submit(
fn=handle_chat_and_clips,
inputs=[chat_input, chatbot],
outputs=[chat_input, chatbot, clips_data_state]
)
with gr.Tab("π¬ Multimodal Chat with Clips"):
# Manual chat layout for full control
chatbot = gr.Chatbot(
[],
type="messages",
label="Video Search Assistant",
height=500,
avatar_images=(None, "https://seeklogo.com/images/O/openai-logo-8284262873-seeklogo.com.png")
)
with gr.Row():
chat_input = gr.Textbox(
show_label=False,
placeholder="Ask me to find clips about cooking...",
lines=1,
scale=4,
container=False,
)
submit_btn = gr.Button("π Search", variant="primary", scale=1, min_width=150)
gr.Examples(
[
"search for clips about the number of computations in llms",
"search for meeting discussions",
"show me sports highlights",
"find outdoor activities"
],
inputs=chat_input,
label="Quick-search examples"
)
gr.Markdown("### π¬ Found Clips")
# State to store clips data for rendering
clips_data_state = gr.State([])
def handle_chat_and_clips(user_message, history):
"""Event handler for chat submission to update chat and clips."""
new_history = chat_agent_mm(user_message, history)
# print(new_history)
clips_data = get_latest_clips_for_display()
return "", new_history, clips_data
# Dynamic clip display
@gr.render(inputs=[clips_data_state])
def show_clips_in_chat(clip_data):
with gr.Column():
if not clip_data:
gr.Markdown("*No clips found yet. Ask the assistant to search for something!*")
return
gr.Markdown(f"**Found {len(clip_data)} relevant clips:**")
for i, clip in enumerate(clip_data):
# Use a column for each clip block to keep them separate
with gr.Column(variant='panel'):
with gr.Row():
with gr.Column(scale=3):
# Clip info
info = clip['info']
gr.Markdown(f"**Clip {info['clip_id']}** from *{info['video_name']}*")
gr.Markdown(f"β±οΈ {info['start_time']:.1f}s - {info['end_time']:.1f}s | π― Relevance: {info['relevance']:.2f}")
# Summary (shortened for chat view)
summary_text = info['summary'][:150] + "..." if len(info['summary']) > 150 else info['summary']
gr.Markdown(f"π {summary_text}")
with gr.Column(scale=1):
# Video player (smaller for chat view)
if clip['video'] and os.path.exists(clip['video']):
gr.Video(clip['video'], label="", height=180, show_label=False)
else:
gr.Markdown("β οΈ *Video not available*")
# Wire up submission events
submit_btn.click(
fn=handle_chat_and_clips,
inputs=[chat_input, chatbot],
outputs=[chat_input, chatbot, clips_data_state]
)
chat_input.submit(
fn=handle_chat_and_clips,
inputs=[chat_input, chatbot],
outputs=[chat_input, chatbot, clips_data_state]
)
# Standalone search interface (keep the original for manual searching)
with gr.Tab("π Manual Search"):
with gr.Row():
with gr.Column(scale=2):
search_input = gr.Textbox(
label="Search Query",
placeholder="Enter your search query (e.g., 'cooking scenes', 'meeting discussions')",
lines=2
)
search_btn = gr.Button("Search Clips", variant="primary")
# Quick search examples
gr.Markdown("**Quick Examples:**")
example_buttons = []
examples = [
"cooking scenes",
"meeting discussions",
"sports highlights",
"outdoor activities"
]
with gr.Row():
for example in examples:
btn = gr.Button(example, size="sm")
example_buttons.append(btn)
btn.click(fn=lambda x=example: x, outputs=search_input)
with gr.Column(scale=1):
search_results_text = gr.Textbox(
label="Search Results Summary",
lines=10,
max_lines=15,
interactive=False
)
# Clips display section
gr.Markdown("## π¬ Found Clips")
clips_display = gr.Column(visible=False)
# Store for clip data
clips_state = gr.State([])
def update_clips_display_manual(query):
if not query.strip():
return "Please enter a search query.", gr.Column(visible=False), []
formatted_results, videos, thumbnails, clip_info = search_and_display_clips(query)
if not videos or all(v is None for v in videos):
return formatted_results, gr.Column(visible=False), []
# Create clip display components
clip_components = []
for i, (video, thumbnail, info) in enumerate(zip(videos, thumbnails, clip_info)):
if video or thumbnail: # Only show if we have media
clip_components.append({
'video': video,
'thumbnail': thumbnail,
'info': info
})
return formatted_results, gr.Column(visible=len(clip_components) > 0), clip_components
# Update the display when search is triggered
search_btn.click(
fn=update_clips_display_manual,
inputs=[search_input],
outputs=[search_results_text, clips_display, clips_state]
)
# Dynamic clip display
@gr.render(inputs=[clips_state])
def show_clips(clip_data):
if not clip_data:
return
for i, clip in enumerate(clip_data):
with gr.Row():
with gr.Column():
# Clip info
info = clip['info']
gr.Markdown(f"### Clip {info['clip_id']} from {info['video_name']}")
gr.Markdown(f"**Time:** {info['start_time']}s - {info['end_time']}s")
gr.Markdown(f"**Relevance Score:** {info['relevance']:.2f}")
# Summary
summary_text = info['summary'][:300] + "..." if len(info['summary']) > 300 else info['summary']
gr.Markdown(f"**Summary:** {summary_text}")
# Video player
if clip['video'] and os.path.exists(clip['video']):
gr.Video(clip['video'], label="Play Clip")
else:
gr.Markdown("β οΈ Video file not available")
gr.Markdown("---")
# Video analyzer tool section
with gr.Tab("πΉ Video Analyzer"):
gr.Markdown("""
**To analyze new videos:**
1. Upload your video file using the interface below
2. Click "Analyze Video" to process the video
3. The processed clips will be automatically added to the searchable database
""")
with gr.Row():
video_file = gr.File(
label="Upload Video",
file_types=[".mp4"],
type="filepath"
)
analyze_btn = gr.Button("Analyze Video", variant="primary")
analysis_output = gr.Textbox(
label="Analysis Status",
lines=5,
interactive=False
)
def analyze_video_local(file_obj):
if not file_obj:
return "Please select a video file first."
try:
# Save uploaded file to a temp file with the same name as the uploaded file
if hasattr(file_obj, 'name'):
original_filename = os.path.basename(file_obj.name)
else:
original_filename = "uploaded_video.mp4"
temp_dir = tempfile.mkdtemp()
tmp_path = os.path.join(temp_dir, original_filename)
with open(tmp_path, "wb") as f:
f.write(file_obj)
# Run the video processing pipeline
run_pipeline(tmp_path)
# Clean up temp file after processing
try:
os.remove(tmp_path)
shutil.rmtree(temp_dir)
except Exception:
pass
return f"β
Video analysis complete for '{original_filename}'. You can now search for clips from this video."
except Exception as e:
return f"β Error during video analysis: {str(e)}"
analyze_btn.click(
fn=analyze_video_local,
inputs=[video_file],
outputs=[analysis_output]
)
# Launch the application
if __name__ == "__main__":
print("π Starting Video Search Agent...")
demo.launch(
server_name="0.0.0.0",
server_port=7860,
share=False,
) |