Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,10 +4,10 @@ import logging
|
|
| 4 |
import os
|
| 5 |
import sys
|
| 6 |
import threading
|
|
|
|
| 7 |
from collections import Counter
|
| 8 |
from io import BytesIO
|
| 9 |
from typing import Dict, List, Optional, Tuple, Union
|
| 10 |
-
|
| 11 |
import gradio as gr
|
| 12 |
import pandas as pd
|
| 13 |
import requests
|
|
@@ -22,17 +22,19 @@ from transformers import (
|
|
| 22 |
DetrImageProcessor,
|
| 23 |
YolosForObjectDetection,
|
| 24 |
YolosImageProcessor,
|
|
|
|
|
|
|
| 25 |
)
|
| 26 |
import nest_asyncio
|
| 27 |
-
|
| 28 |
# ------------------------------
|
| 29 |
# Configuration
|
| 30 |
# ------------------------------
|
| 31 |
-
|
| 32 |
# Configure logging for debugging and monitoring
|
| 33 |
logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
|
| 34 |
logger = logging.getLogger(__name__)
|
| 35 |
-
|
|
|
|
|
|
|
| 36 |
# Define constants for model and server configuration
|
| 37 |
CONFIDENCE_THRESHOLD: float = 0.5 # Default threshold for object detection confidence
|
| 38 |
VALID_MODELS: List[str] = [
|
|
@@ -41,7 +43,9 @@ VALID_MODELS: List[str] = [
|
|
| 41 |
"facebook/detr-resnet-50-panoptic",
|
| 42 |
"facebook/detr-resnet-101-panoptic",
|
| 43 |
"hustvl/yolos-tiny",
|
|
|
|
| 44 |
"hustvl/yolos-base",
|
|
|
|
| 45 |
]
|
| 46 |
MODEL_DESCRIPTIONS: Dict[str, str] = {
|
| 47 |
"facebook/detr-resnet-50": "DETR with ResNet-50 for object detection. Fast and accurate.",
|
|
@@ -49,22 +53,21 @@ MODEL_DESCRIPTIONS: Dict[str, str] = {
|
|
| 49 |
"facebook/detr-resnet-50-panoptic": "DETR with ResNet-50 for panoptic segmentation.",
|
| 50 |
"facebook/detr-resnet-101-panoptic": "DETR with ResNet-101 for panoptic segmentation.",
|
| 51 |
"hustvl/yolos-tiny": "YOLOS Tiny. Lightweight and fast.",
|
| 52 |
-
"hustvl/yolos-
|
|
|
|
|
|
|
| 53 |
}
|
| 54 |
DEFAULT_GRADIO_PORT: int = 7860 # Default port for Gradio UI
|
| 55 |
DEFAULT_FASTAPI_PORT: int = 8000 # Default port for FastAPI server
|
| 56 |
PORT_RANGE: range = range(7860, 7870) # Range of ports to try for Gradio
|
| 57 |
MAX_PORT_ATTEMPTS: int = 10 # Maximum attempts to find an available port
|
| 58 |
-
|
| 59 |
# Thread-safe storage for lazy-loaded models and processors
|
| 60 |
models: Dict[str, any] = {}
|
| 61 |
processors: Dict[str, any] = {}
|
| 62 |
model_lock = threading.Lock()
|
| 63 |
-
|
| 64 |
# ------------------------------
|
| 65 |
# Image Processing
|
| 66 |
# ------------------------------
|
| 67 |
-
|
| 68 |
def process_image(
|
| 69 |
image: Optional[Image.Image],
|
| 70 |
url: Optional[str],
|
|
@@ -74,14 +77,12 @@ def process_image(
|
|
| 74 |
) -> Union[Dict, Tuple[Optional[Image.Image], Optional[pd.DataFrame], Optional[pd.DataFrame], Optional[pd.DataFrame], str]]:
|
| 75 |
"""
|
| 76 |
Process an image for object detection or panoptic segmentation, handling Gradio and FastAPI inputs.
|
| 77 |
-
|
| 78 |
Args:
|
| 79 |
image: PIL Image object from file upload (optional).
|
| 80 |
url: URL of the image to process (optional).
|
| 81 |
model_name: Name of the model to use (must be in VALID_MODELS).
|
| 82 |
for_json: If True, return JSON dict for API/JSON tab; else, return tuple for Gradio Home tab.
|
| 83 |
confidence_threshold: Minimum confidence score for detection (default: 0.5).
|
| 84 |
-
|
| 85 |
Returns:
|
| 86 |
For JSON: Dict with base64-encoded image, detected objects, and confidence scores.
|
| 87 |
For Gradio: Tuple of (annotated image, objects DataFrame, unique objects DataFrame, properties DataFrame, error message).
|
|
@@ -95,16 +96,17 @@ def process_image(
|
|
| 95 |
if model_name not in VALID_MODELS:
|
| 96 |
error_msg = f"Invalid model: {model_name}. Choose from: {VALID_MODELS}"
|
| 97 |
return {"error": error_msg} if for_json else (None, None, None, None, error_msg)
|
| 98 |
-
|
|
|
|
|
|
|
|
|
|
| 99 |
# Calculate margin threshold: (1 - confidence_threshold) / 2 + confidence_threshold
|
| 100 |
margin_threshold = (1 - confidence_threshold) / 2 + confidence_threshold
|
| 101 |
-
|
| 102 |
# Load image from URL if provided
|
| 103 |
if url:
|
| 104 |
response = requests.get(url, timeout=10)
|
| 105 |
response.raise_for_status()
|
| 106 |
image = Image.open(BytesIO(response.content)).convert("RGB")
|
| 107 |
-
|
| 108 |
# Load model and processor thread-safely
|
| 109 |
with model_lock:
|
| 110 |
if model_name not in models:
|
|
@@ -112,32 +114,35 @@ def process_image(
|
|
| 112 |
try:
|
| 113 |
# Select appropriate model and processor based on model name
|
| 114 |
if "yolos" in model_name:
|
| 115 |
-
models[model_name] = YolosForObjectDetection.from_pretrained(model_name)
|
| 116 |
processors[model_name] = YolosImageProcessor.from_pretrained(model_name)
|
| 117 |
elif "panoptic" in model_name:
|
| 118 |
-
models[model_name] = DetrForSegmentation.from_pretrained(model_name)
|
| 119 |
processors[model_name] = DetrImageProcessor.from_pretrained(model_name)
|
|
|
|
|
|
|
|
|
|
| 120 |
else:
|
| 121 |
-
models[model_name] = DetrForObjectDetection.from_pretrained(model_name)
|
| 122 |
processors[model_name] = DetrImageProcessor.from_pretrained(model_name)
|
| 123 |
except Exception as e:
|
| 124 |
error_msg = f"Failed to load model: {str(e)}"
|
| 125 |
logger.error(error_msg)
|
| 126 |
return {"error": error_msg} if for_json else (None, None, None, None, error_msg)
|
| 127 |
model, processor = models[model_name], processors[model_name]
|
| 128 |
-
|
| 129 |
# Prepare image for model processing
|
| 130 |
inputs = processor(images=image, return_tensors="pt")
|
|
|
|
|
|
|
|
|
|
| 131 |
with torch.no_grad():
|
| 132 |
outputs = model(**inputs)
|
| 133 |
-
|
| 134 |
# Initialize drawing context for annotations
|
| 135 |
draw = ImageDraw.Draw(image)
|
| 136 |
object_names: List[str] = []
|
| 137 |
confidence_scores: List[float] = []
|
| 138 |
object_counter = Counter()
|
| 139 |
target_sizes = torch.tensor([image.size[::-1]])
|
| 140 |
-
|
| 141 |
# Process results based on model type (panoptic or object detection)
|
| 142 |
if "panoptic" in model_name:
|
| 143 |
# Handle panoptic segmentation
|
|
@@ -161,29 +166,26 @@ def process_image(
|
|
| 161 |
if score > confidence_threshold:
|
| 162 |
object_names.append(label_name)
|
| 163 |
confidence_scores.append(float(score))
|
| 164 |
-
object_counter[label_name] = float(score)
|
| 165 |
else:
|
| 166 |
# Handle object detection
|
| 167 |
-
results = processor.post_process_object_detection(outputs, target_sizes=target_sizes)[0]
|
| 168 |
for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
object_counter[label_name] = float(score)
|
| 182 |
-
|
| 183 |
# Compile unique objects and their highest confidence scores
|
| 184 |
unique_objects = list(object_counter.keys())
|
| 185 |
unique_confidences = [object_counter[obj] for obj in unique_objects]
|
| 186 |
-
|
| 187 |
# Calculate image properties (metadata)
|
| 188 |
properties: Dict[str, str] = {
|
| 189 |
"Format": image.format if hasattr(image, "format") and image.format else "Unknown",
|
|
@@ -207,7 +209,6 @@ def process_image(
|
|
| 207 |
properties["StdDev (R,G,B)"] = ", ".join(f"{s:.2f}" for s in stat.stddev)
|
| 208 |
except Exception as e:
|
| 209 |
logger.error(f"Error calculating image stats: {str(e)}")
|
| 210 |
-
|
| 211 |
# Prepare output based on request type
|
| 212 |
if for_json:
|
| 213 |
# Return JSON with base64-encoded image
|
|
@@ -231,9 +232,8 @@ def process_image(
|
|
| 231 |
pd.DataFrame({"Unique Object": unique_objects, "Confidence Score": [f"{score:.2f}" for score in unique_confidences]})
|
| 232 |
if unique_objects else pd.DataFrame(columns=["Unique Object", "Confidence Score"])
|
| 233 |
)
|
| 234 |
-
properties_df = pd.DataFrame([properties]) if properties else pd.DataFrame(columns=properties.keys())
|
| 235 |
return image, objects_df, unique_objects_df, properties_df, ""
|
| 236 |
-
|
| 237 |
except requests.RequestException as e:
|
| 238 |
# Handle URL fetch errors
|
| 239 |
error_msg = f"Error fetching image from URL: {str(e)}"
|
|
@@ -244,13 +244,10 @@ def process_image(
|
|
| 244 |
error_msg = f"Error processing image: {str(e)}"
|
| 245 |
logger.error(f"{error_msg}\n{traceback.format_exc()}")
|
| 246 |
return {"error": error_msg} if for_json else (None, None, None, None, error_msg)
|
| 247 |
-
|
| 248 |
# ------------------------------
|
| 249 |
# FastAPI Setup
|
| 250 |
# ------------------------------
|
| 251 |
-
|
| 252 |
app = FastAPI(title="Object Detection API")
|
| 253 |
-
|
| 254 |
@app.post("/detect")
|
| 255 |
async def detect_objects_endpoint(
|
| 256 |
file: Optional[UploadFile] = File(None),
|
|
@@ -260,16 +257,13 @@ async def detect_objects_endpoint(
|
|
| 260 |
) -> JSONResponse:
|
| 261 |
"""
|
| 262 |
FastAPI endpoint to detect objects in an image from file upload or URL.
|
| 263 |
-
|
| 264 |
Args:
|
| 265 |
file: Uploaded image file (optional).
|
| 266 |
image_url: URL of the image (optional).
|
| 267 |
model_name: Model to use for detection (default: first VALID_MODELS entry).
|
| 268 |
confidence_threshold: Confidence threshold for detection (default: 0.5).
|
| 269 |
-
|
| 270 |
Returns:
|
| 271 |
JSONResponse with base64-encoded image, detected objects, and confidence scores.
|
| 272 |
-
|
| 273 |
Raises:
|
| 274 |
HTTPException: For invalid inputs or processing errors.
|
| 275 |
"""
|
|
@@ -277,9 +271,6 @@ async def detect_objects_endpoint(
|
|
| 277 |
# Validate input: ensure exactly one of file or URL
|
| 278 |
if (file is None and not image_url) or (file is not None and image_url):
|
| 279 |
raise HTTPException(status_code=400, detail="Provide either an image file or an image URL, not both.")
|
| 280 |
-
# Validate confidence threshold
|
| 281 |
-
if not 0 <= confidence_threshold <= 1:
|
| 282 |
-
raise HTTPException(status_code=400, detail="Confidence threshold must be between 0 and 1.")
|
| 283 |
# Load image from file if provided
|
| 284 |
image = None
|
| 285 |
if file:
|
|
@@ -297,18 +288,14 @@ async def detect_objects_endpoint(
|
|
| 297 |
except Exception as e:
|
| 298 |
logger.error(f"Error in FastAPI endpoint: {str(e)}\n{traceback.format_exc()}")
|
| 299 |
raise HTTPException(status_code=500, detail=f"Error processing image: {str(e)}")
|
| 300 |
-
|
| 301 |
# ------------------------------
|
| 302 |
# Gradio UI Setup
|
| 303 |
# ------------------------------
|
| 304 |
-
|
| 305 |
def create_gradio_ui() -> gr.Blocks:
|
| 306 |
"""
|
| 307 |
Create and configure the Gradio UI for object detection with Home, JSON, and Help tabs.
|
| 308 |
-
|
| 309 |
Returns:
|
| 310 |
Gradio Blocks object representing the UI.
|
| 311 |
-
|
| 312 |
Raises:
|
| 313 |
RuntimeError: If UI creation fails.
|
| 314 |
"""
|
|
@@ -319,11 +306,11 @@ def create_gradio_ui() -> gr.Blocks:
|
|
| 319 |
gr.Markdown(
|
| 320 |
f"""
|
| 321 |
# π Object Detection App
|
| 322 |
-
Upload an image or provide a URL to detect objects using transformer models (DETR, YOLOS).
|
| 323 |
Running on port: {os.getenv('GRADIO_SERVER_PORT', 'auto-selected')}
|
|
|
|
| 324 |
"""
|
| 325 |
)
|
| 326 |
-
|
| 327 |
# Create tabbed interface
|
| 328 |
with gr.Tabs():
|
| 329 |
# Home tab (formerly Image Upload)
|
|
@@ -335,6 +322,8 @@ def create_gradio_ui() -> gr.Blocks:
|
|
| 335 |
# Model selection dropdown
|
| 336 |
model_choice = gr.Dropdown(choices=VALID_MODELS, value=VALID_MODELS[0], label="π Select Model")
|
| 337 |
model_info = gr.Markdown(f"**Model Info**: {MODEL_DESCRIPTIONS[VALID_MODELS[0]]}")
|
|
|
|
|
|
|
| 338 |
# Image upload input
|
| 339 |
image_input = gr.Image(type="pil", label="π· Upload Image")
|
| 340 |
# Image URL input
|
|
@@ -343,14 +332,12 @@ def create_gradio_ui() -> gr.Blocks:
|
|
| 343 |
with gr.Row():
|
| 344 |
submit_btn = gr.Button("β¨ Detect", variant="primary")
|
| 345 |
clear_btn = gr.Button("ποΈ Clear", variant="secondary")
|
| 346 |
-
|
| 347 |
# Update model info when model changes
|
| 348 |
model_choice.change(
|
| 349 |
fn=lambda model_name: f"**Model Info**: {MODEL_DESCRIPTIONS.get(model_name, 'No description available.')}",
|
| 350 |
inputs=model_choice,
|
| 351 |
outputs=model_info,
|
| 352 |
)
|
| 353 |
-
|
| 354 |
# Right column for results
|
| 355 |
with gr.Column(scale=2):
|
| 356 |
gr.Markdown("### Results")
|
|
@@ -364,21 +351,18 @@ def create_gradio_ui() -> gr.Blocks:
|
|
| 364 |
unique_objects_output = gr.DataFrame(label="π Unique Objects", interactive=False)
|
| 365 |
# Image properties table
|
| 366 |
properties_output = gr.DataFrame(label="π Image Properties", interactive=False)
|
| 367 |
-
|
| 368 |
# Process image when Detect button is clicked
|
| 369 |
submit_btn.click(
|
| 370 |
fn=process_image,
|
| 371 |
-
inputs=[image_input, image_url_input, model_choice],
|
| 372 |
outputs=[output_image, objects_output, unique_objects_output, properties_output, error_output],
|
| 373 |
)
|
| 374 |
-
|
| 375 |
# Clear all inputs and outputs
|
| 376 |
clear_btn.click(
|
| 377 |
-
fn=lambda: [None, "",
|
| 378 |
inputs=None,
|
| 379 |
-
outputs=[image_input, image_url_input, output_image, objects_output, unique_objects_output, properties_output, error_output],
|
| 380 |
)
|
| 381 |
-
|
| 382 |
# JSON tab for API-like output
|
| 383 |
with gr.Tab("π JSON"):
|
| 384 |
with gr.Row():
|
|
@@ -388,64 +372,64 @@ def create_gradio_ui() -> gr.Blocks:
|
|
| 388 |
# Model selection dropdown
|
| 389 |
url_model_choice = gr.Dropdown(choices=VALID_MODELS, value=VALID_MODELS[0], label="π Select Model")
|
| 390 |
url_model_info = gr.Markdown(f"**Model Info**: {MODEL_DESCRIPTIONS[VALID_MODELS[0]]}")
|
|
|
|
|
|
|
| 391 |
# Image upload input
|
| 392 |
image_input_json = gr.Image(type="pil", label="π· Upload Image")
|
| 393 |
# Image URL input
|
| 394 |
image_url_input_json = gr.Textbox(label="π Image URL", placeholder="https://example.com/image.jpg")
|
| 395 |
# Process button
|
| 396 |
url_submit_btn = gr.Button("π Process", variant="primary")
|
| 397 |
-
|
| 398 |
# Update model info when model changes
|
| 399 |
url_model_choice.change(
|
| 400 |
fn=lambda model_name: f"**Model Info**: {MODEL_DESCRIPTIONS.get(model_name, 'No description available.')}",
|
| 401 |
inputs=url_model_choice,
|
| 402 |
outputs=url_model_info,
|
| 403 |
)
|
| 404 |
-
|
| 405 |
# Right column for JSON output
|
| 406 |
with gr.Column(scale=1):
|
| 407 |
# JSON output display
|
| 408 |
url_output = gr.JSON(label="API Response")
|
| 409 |
-
|
| 410 |
# Process image and return JSON when Process button is clicked
|
| 411 |
url_submit_btn.click(
|
| 412 |
-
fn=lambda img, url, model: process_image(img, url, model, for_json=True),
|
| 413 |
-
inputs=[image_input_json, image_url_input_json, url_model_choice],
|
| 414 |
outputs=[url_output],
|
| 415 |
)
|
| 416 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 417 |
# Help tab with usage instructions
|
| 418 |
with gr.Tab("βΉοΈ Help"):
|
| 419 |
gr.Markdown(
|
| 420 |
"""
|
| 421 |
## How to Use
|
| 422 |
-
- **Home**: Select a model, upload an image or provide a URL, click "Detect" to see results.
|
| 423 |
-
- **JSON**: Select a model, upload an image or enter a URL, click "Process" for JSON output.
|
| 424 |
-
- **Models**: Choose DETR (detection or panoptic)
|
| 425 |
-
- **Clear**: Reset inputs/outputs in Home tab.
|
| 426 |
- **Errors**: Check error box (Home) or JSON response (JSON) for issues.
|
| 427 |
-
|
| 428 |
## Tips
|
| 429 |
- Use high-quality images for better results.
|
| 430 |
- Panoptic models provide segmentation masks for complex scenes.
|
| 431 |
- YOLOS-Tiny is faster for resource-constrained devices.
|
|
|
|
| 432 |
"""
|
| 433 |
)
|
| 434 |
-
|
| 435 |
return demo
|
| 436 |
-
|
| 437 |
except Exception as e:
|
| 438 |
logger.error(f"Error creating Gradio UI: {str(e)}\n{traceback.format_exc()}")
|
| 439 |
raise RuntimeError(f"Failed to create Gradio UI: {str(e)}")
|
| 440 |
-
|
| 441 |
# ------------------------------
|
| 442 |
# Launcher
|
| 443 |
# ------------------------------
|
| 444 |
-
|
| 445 |
def parse_args() -> argparse.Namespace:
|
| 446 |
"""
|
| 447 |
Parse command-line arguments for configuring the application.
|
| 448 |
-
|
| 449 |
Returns:
|
| 450 |
Parsed arguments as a Namespace object.
|
| 451 |
"""
|
|
@@ -464,16 +448,13 @@ def parse_args() -> argparse.Namespace:
|
|
| 464 |
if not 0 <= args.confidence_threshold <= 1:
|
| 465 |
parser.error("Confidence threshold must be between 0 and 1.")
|
| 466 |
return args
|
| 467 |
-
|
| 468 |
def find_available_port(start_port: int, port_range: range, max_attempts: int) -> Optional[int]:
|
| 469 |
"""
|
| 470 |
Find an available port within the specified range.
|
| 471 |
-
|
| 472 |
Args:
|
| 473 |
start_port: Initial port to try.
|
| 474 |
port_range: Range of ports to attempt.
|
| 475 |
max_attempts: Maximum number of ports to try.
|
| 476 |
-
|
| 477 |
Returns:
|
| 478 |
Available port number, or None if no port is found.
|
| 479 |
"""
|
|
@@ -497,11 +478,9 @@ def find_available_port(start_port: int, port_range: range, max_attempts: int) -
|
|
| 497 |
raise
|
| 498 |
logger.error(f"No available port in range {min(port_range)}-{max(port_range)}")
|
| 499 |
return None
|
| 500 |
-
|
| 501 |
def main() -> None:
|
| 502 |
"""
|
| 503 |
Launch the Gradio UI and optional FastAPI server.
|
| 504 |
-
|
| 505 |
Raises:
|
| 506 |
SystemExit: On interruption or critical errors.
|
| 507 |
"""
|
|
@@ -516,7 +495,6 @@ def main() -> None:
|
|
| 516 |
if gradio_port is None:
|
| 517 |
logger.error("Failed to find an available port for Gradio UI")
|
| 518 |
sys.exit(1)
|
| 519 |
-
|
| 520 |
# Start FastAPI server in a thread if enabled
|
| 521 |
if args.enable_fastapi:
|
| 522 |
logger.info(f"Starting FastAPI on port {args.fastapi_port}")
|
|
@@ -525,18 +503,15 @@ def main() -> None:
|
|
| 525 |
daemon=True
|
| 526 |
)
|
| 527 |
fastapi_thread.start()
|
| 528 |
-
|
| 529 |
# Launch Gradio UI
|
| 530 |
logger.info(f"Starting Gradio UI on port {gradio_port}")
|
| 531 |
demo = create_gradio_ui()
|
| 532 |
demo.launch(server_port=gradio_port, server_name="0.0.0.0")
|
| 533 |
-
|
| 534 |
except KeyboardInterrupt:
|
| 535 |
logger.info("Application terminated by user.")
|
| 536 |
sys.exit(0)
|
| 537 |
except Exception as e:
|
| 538 |
logger.error(f"Error: {str(e)}\n{traceback.format_exc()}")
|
| 539 |
sys.exit(1)
|
| 540 |
-
|
| 541 |
if __name__ == "__main__":
|
| 542 |
main()
|
|
|
|
| 4 |
import os
|
| 5 |
import sys
|
| 6 |
import threading
|
| 7 |
+
import traceback
|
| 8 |
from collections import Counter
|
| 9 |
from io import BytesIO
|
| 10 |
from typing import Dict, List, Optional, Tuple, Union
|
|
|
|
| 11 |
import gradio as gr
|
| 12 |
import pandas as pd
|
| 13 |
import requests
|
|
|
|
| 22 |
DetrImageProcessor,
|
| 23 |
YolosForObjectDetection,
|
| 24 |
YolosImageProcessor,
|
| 25 |
+
DeformableDetrForObjectDetection,
|
| 26 |
+
DeformableDetrImageProcessor,
|
| 27 |
)
|
| 28 |
import nest_asyncio
|
|
|
|
| 29 |
# ------------------------------
|
| 30 |
# Configuration
|
| 31 |
# ------------------------------
|
|
|
|
| 32 |
# Configure logging for debugging and monitoring
|
| 33 |
logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
|
| 34 |
logger = logging.getLogger(__name__)
|
| 35 |
+
# Define device
|
| 36 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 37 |
+
logger.info(f"Using device: {device}")
|
| 38 |
# Define constants for model and server configuration
|
| 39 |
CONFIDENCE_THRESHOLD: float = 0.5 # Default threshold for object detection confidence
|
| 40 |
VALID_MODELS: List[str] = [
|
|
|
|
| 43 |
"facebook/detr-resnet-50-panoptic",
|
| 44 |
"facebook/detr-resnet-101-panoptic",
|
| 45 |
"hustvl/yolos-tiny",
|
| 46 |
+
"hustvl/yolos-small",
|
| 47 |
"hustvl/yolos-base",
|
| 48 |
+
"SenseTime/deformable-detr",
|
| 49 |
]
|
| 50 |
MODEL_DESCRIPTIONS: Dict[str, str] = {
|
| 51 |
"facebook/detr-resnet-50": "DETR with ResNet-50 for object detection. Fast and accurate.",
|
|
|
|
| 53 |
"facebook/detr-resnet-50-panoptic": "DETR with ResNet-50 for panoptic segmentation.",
|
| 54 |
"facebook/detr-resnet-101-panoptic": "DETR with ResNet-101 for panoptic segmentation.",
|
| 55 |
"hustvl/yolos-tiny": "YOLOS Tiny. Lightweight and fast.",
|
| 56 |
+
"hustvl/yolos-small": "YOLOS Small. Medium speed and accuracy.",
|
| 57 |
+
"hustvl/yolos-base": "YOLOS Base. Balances speed and accuracy.",
|
| 58 |
+
"SenseTime/deformable-detr": "Deformable DETR. Improved efficiency with deformable attention.",
|
| 59 |
}
|
| 60 |
DEFAULT_GRADIO_PORT: int = 7860 # Default port for Gradio UI
|
| 61 |
DEFAULT_FASTAPI_PORT: int = 8000 # Default port for FastAPI server
|
| 62 |
PORT_RANGE: range = range(7860, 7870) # Range of ports to try for Gradio
|
| 63 |
MAX_PORT_ATTEMPTS: int = 10 # Maximum attempts to find an available port
|
|
|
|
| 64 |
# Thread-safe storage for lazy-loaded models and processors
|
| 65 |
models: Dict[str, any] = {}
|
| 66 |
processors: Dict[str, any] = {}
|
| 67 |
model_lock = threading.Lock()
|
|
|
|
| 68 |
# ------------------------------
|
| 69 |
# Image Processing
|
| 70 |
# ------------------------------
|
|
|
|
| 71 |
def process_image(
|
| 72 |
image: Optional[Image.Image],
|
| 73 |
url: Optional[str],
|
|
|
|
| 77 |
) -> Union[Dict, Tuple[Optional[Image.Image], Optional[pd.DataFrame], Optional[pd.DataFrame], Optional[pd.DataFrame], str]]:
|
| 78 |
"""
|
| 79 |
Process an image for object detection or panoptic segmentation, handling Gradio and FastAPI inputs.
|
|
|
|
| 80 |
Args:
|
| 81 |
image: PIL Image object from file upload (optional).
|
| 82 |
url: URL of the image to process (optional).
|
| 83 |
model_name: Name of the model to use (must be in VALID_MODELS).
|
| 84 |
for_json: If True, return JSON dict for API/JSON tab; else, return tuple for Gradio Home tab.
|
| 85 |
confidence_threshold: Minimum confidence score for detection (default: 0.5).
|
|
|
|
| 86 |
Returns:
|
| 87 |
For JSON: Dict with base64-encoded image, detected objects, and confidence scores.
|
| 88 |
For Gradio: Tuple of (annotated image, objects DataFrame, unique objects DataFrame, properties DataFrame, error message).
|
|
|
|
| 96 |
if model_name not in VALID_MODELS:
|
| 97 |
error_msg = f"Invalid model: {model_name}. Choose from: {VALID_MODELS}"
|
| 98 |
return {"error": error_msg} if for_json else (None, None, None, None, error_msg)
|
| 99 |
+
# Validate confidence threshold
|
| 100 |
+
if not 0 <= confidence_threshold <= 1:
|
| 101 |
+
error_msg = "Confidence threshold must be between 0 and 1."
|
| 102 |
+
return {"error": error_msg} if for_json else (None, None, None, None, error_msg)
|
| 103 |
# Calculate margin threshold: (1 - confidence_threshold) / 2 + confidence_threshold
|
| 104 |
margin_threshold = (1 - confidence_threshold) / 2 + confidence_threshold
|
|
|
|
| 105 |
# Load image from URL if provided
|
| 106 |
if url:
|
| 107 |
response = requests.get(url, timeout=10)
|
| 108 |
response.raise_for_status()
|
| 109 |
image = Image.open(BytesIO(response.content)).convert("RGB")
|
|
|
|
| 110 |
# Load model and processor thread-safely
|
| 111 |
with model_lock:
|
| 112 |
if model_name not in models:
|
|
|
|
| 114 |
try:
|
| 115 |
# Select appropriate model and processor based on model name
|
| 116 |
if "yolos" in model_name:
|
| 117 |
+
models[model_name] = YolosForObjectDetection.from_pretrained(model_name).eval().to(device)
|
| 118 |
processors[model_name] = YolosImageProcessor.from_pretrained(model_name)
|
| 119 |
elif "panoptic" in model_name:
|
| 120 |
+
models[model_name] = DetrForSegmentation.from_pretrained(model_name).eval().to(device)
|
| 121 |
processors[model_name] = DetrImageProcessor.from_pretrained(model_name)
|
| 122 |
+
elif "deformable" in model_name:
|
| 123 |
+
models[model_name] = DeformableDetrForObjectDetection.from_pretrained(model_name).eval().to(device)
|
| 124 |
+
processors[model_name] = DeformableDetrImageProcessor.from_pretrained(model_name)
|
| 125 |
else:
|
| 126 |
+
models[model_name] = DetrForObjectDetection.from_pretrained(model_name).eval().to(device)
|
| 127 |
processors[model_name] = DetrImageProcessor.from_pretrained(model_name)
|
| 128 |
except Exception as e:
|
| 129 |
error_msg = f"Failed to load model: {str(e)}"
|
| 130 |
logger.error(error_msg)
|
| 131 |
return {"error": error_msg} if for_json else (None, None, None, None, error_msg)
|
| 132 |
model, processor = models[model_name], processors[model_name]
|
|
|
|
| 133 |
# Prepare image for model processing
|
| 134 |
inputs = processor(images=image, return_tensors="pt")
|
| 135 |
+
# Move inputs to device if using GPU
|
| 136 |
+
if device == "cuda":
|
| 137 |
+
inputs = {k: v.to(device) for k, v in inputs.items()}
|
| 138 |
with torch.no_grad():
|
| 139 |
outputs = model(**inputs)
|
|
|
|
| 140 |
# Initialize drawing context for annotations
|
| 141 |
draw = ImageDraw.Draw(image)
|
| 142 |
object_names: List[str] = []
|
| 143 |
confidence_scores: List[float] = []
|
| 144 |
object_counter = Counter()
|
| 145 |
target_sizes = torch.tensor([image.size[::-1]])
|
|
|
|
| 146 |
# Process results based on model type (panoptic or object detection)
|
| 147 |
if "panoptic" in model_name:
|
| 148 |
# Handle panoptic segmentation
|
|
|
|
| 166 |
if score > confidence_threshold:
|
| 167 |
object_names.append(label_name)
|
| 168 |
confidence_scores.append(float(score))
|
| 169 |
+
object_counter[label_name] = max(object_counter.get(label_name, 0.0), float(score))
|
| 170 |
else:
|
| 171 |
# Handle object detection
|
| 172 |
+
results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=confidence_threshold)[0]
|
| 173 |
for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
|
| 174 |
+
x, y, x2, y2 = box.tolist()
|
| 175 |
+
label_name = model.config.id2label.get(label.item(), "Unknown")
|
| 176 |
+
text = f"{label_name}: {score:.2f}"
|
| 177 |
+
text_bbox = draw.textbbox((0, 0), text)
|
| 178 |
+
text_width, text_height = text_bbox[2] - text_bbox[0], text_bbox[3] - text_bbox[1]
|
| 179 |
+
# Use yellow for confidence_threshold <= score < margin_threshold, green for >= margin_threshold
|
| 180 |
+
color = "#FFFF00" if score < margin_threshold else "#32CD32"
|
| 181 |
+
draw.rectangle([x, y, x2, y2], outline=color, width=2)
|
| 182 |
+
draw.text((x2 - text_width - 2, y - text_height - 2), text, fill=color)
|
| 183 |
+
object_names.append(label_name)
|
| 184 |
+
confidence_scores.append(float(score))
|
| 185 |
+
object_counter[label_name] = max(object_counter.get(label_name, 0.0), float(score))
|
|
|
|
|
|
|
| 186 |
# Compile unique objects and their highest confidence scores
|
| 187 |
unique_objects = list(object_counter.keys())
|
| 188 |
unique_confidences = [object_counter[obj] for obj in unique_objects]
|
|
|
|
| 189 |
# Calculate image properties (metadata)
|
| 190 |
properties: Dict[str, str] = {
|
| 191 |
"Format": image.format if hasattr(image, "format") and image.format else "Unknown",
|
|
|
|
| 209 |
properties["StdDev (R,G,B)"] = ", ".join(f"{s:.2f}" for s in stat.stddev)
|
| 210 |
except Exception as e:
|
| 211 |
logger.error(f"Error calculating image stats: {str(e)}")
|
|
|
|
| 212 |
# Prepare output based on request type
|
| 213 |
if for_json:
|
| 214 |
# Return JSON with base64-encoded image
|
|
|
|
| 232 |
pd.DataFrame({"Unique Object": unique_objects, "Confidence Score": [f"{score:.2f}" for score in unique_confidences]})
|
| 233 |
if unique_objects else pd.DataFrame(columns=["Unique Object", "Confidence Score"])
|
| 234 |
)
|
| 235 |
+
properties_df = pd.DataFrame([properties]) if properties else pd.DataFrame(columns=list(properties.keys()))
|
| 236 |
return image, objects_df, unique_objects_df, properties_df, ""
|
|
|
|
| 237 |
except requests.RequestException as e:
|
| 238 |
# Handle URL fetch errors
|
| 239 |
error_msg = f"Error fetching image from URL: {str(e)}"
|
|
|
|
| 244 |
error_msg = f"Error processing image: {str(e)}"
|
| 245 |
logger.error(f"{error_msg}\n{traceback.format_exc()}")
|
| 246 |
return {"error": error_msg} if for_json else (None, None, None, None, error_msg)
|
|
|
|
| 247 |
# ------------------------------
|
| 248 |
# FastAPI Setup
|
| 249 |
# ------------------------------
|
|
|
|
| 250 |
app = FastAPI(title="Object Detection API")
|
|
|
|
| 251 |
@app.post("/detect")
|
| 252 |
async def detect_objects_endpoint(
|
| 253 |
file: Optional[UploadFile] = File(None),
|
|
|
|
| 257 |
) -> JSONResponse:
|
| 258 |
"""
|
| 259 |
FastAPI endpoint to detect objects in an image from file upload or URL.
|
|
|
|
| 260 |
Args:
|
| 261 |
file: Uploaded image file (optional).
|
| 262 |
image_url: URL of the image (optional).
|
| 263 |
model_name: Model to use for detection (default: first VALID_MODELS entry).
|
| 264 |
confidence_threshold: Confidence threshold for detection (default: 0.5).
|
|
|
|
| 265 |
Returns:
|
| 266 |
JSONResponse with base64-encoded image, detected objects, and confidence scores.
|
|
|
|
| 267 |
Raises:
|
| 268 |
HTTPException: For invalid inputs or processing errors.
|
| 269 |
"""
|
|
|
|
| 271 |
# Validate input: ensure exactly one of file or URL
|
| 272 |
if (file is None and not image_url) or (file is not None and image_url):
|
| 273 |
raise HTTPException(status_code=400, detail="Provide either an image file or an image URL, not both.")
|
|
|
|
|
|
|
|
|
|
| 274 |
# Load image from file if provided
|
| 275 |
image = None
|
| 276 |
if file:
|
|
|
|
| 288 |
except Exception as e:
|
| 289 |
logger.error(f"Error in FastAPI endpoint: {str(e)}\n{traceback.format_exc()}")
|
| 290 |
raise HTTPException(status_code=500, detail=f"Error processing image: {str(e)}")
|
|
|
|
| 291 |
# ------------------------------
|
| 292 |
# Gradio UI Setup
|
| 293 |
# ------------------------------
|
|
|
|
| 294 |
def create_gradio_ui() -> gr.Blocks:
|
| 295 |
"""
|
| 296 |
Create and configure the Gradio UI for object detection with Home, JSON, and Help tabs.
|
|
|
|
| 297 |
Returns:
|
| 298 |
Gradio Blocks object representing the UI.
|
|
|
|
| 299 |
Raises:
|
| 300 |
RuntimeError: If UI creation fails.
|
| 301 |
"""
|
|
|
|
| 306 |
gr.Markdown(
|
| 307 |
f"""
|
| 308 |
# π Object Detection App
|
| 309 |
+
Upload an image or provide a URL to detect objects using transformer models (DETR, YOLOS, Deformable DETR).
|
| 310 |
Running on port: {os.getenv('GRADIO_SERVER_PORT', 'auto-selected')}
|
| 311 |
+
Device: {device.upper()}
|
| 312 |
"""
|
| 313 |
)
|
|
|
|
| 314 |
# Create tabbed interface
|
| 315 |
with gr.Tabs():
|
| 316 |
# Home tab (formerly Image Upload)
|
|
|
|
| 322 |
# Model selection dropdown
|
| 323 |
model_choice = gr.Dropdown(choices=VALID_MODELS, value=VALID_MODELS[0], label="π Select Model")
|
| 324 |
model_info = gr.Markdown(f"**Model Info**: {MODEL_DESCRIPTIONS[VALID_MODELS[0]]}")
|
| 325 |
+
# Confidence threshold slider
|
| 326 |
+
confidence_slider = gr.Slider(minimum=0, maximum=1, value=0.5, step=0.01, label="Confidence Threshold")
|
| 327 |
# Image upload input
|
| 328 |
image_input = gr.Image(type="pil", label="π· Upload Image")
|
| 329 |
# Image URL input
|
|
|
|
| 332 |
with gr.Row():
|
| 333 |
submit_btn = gr.Button("β¨ Detect", variant="primary")
|
| 334 |
clear_btn = gr.Button("ποΈ Clear", variant="secondary")
|
|
|
|
| 335 |
# Update model info when model changes
|
| 336 |
model_choice.change(
|
| 337 |
fn=lambda model_name: f"**Model Info**: {MODEL_DESCRIPTIONS.get(model_name, 'No description available.')}",
|
| 338 |
inputs=model_choice,
|
| 339 |
outputs=model_info,
|
| 340 |
)
|
|
|
|
| 341 |
# Right column for results
|
| 342 |
with gr.Column(scale=2):
|
| 343 |
gr.Markdown("### Results")
|
|
|
|
| 351 |
unique_objects_output = gr.DataFrame(label="π Unique Objects", interactive=False)
|
| 352 |
# Image properties table
|
| 353 |
properties_output = gr.DataFrame(label="π Image Properties", interactive=False)
|
|
|
|
| 354 |
# Process image when Detect button is clicked
|
| 355 |
submit_btn.click(
|
| 356 |
fn=process_image,
|
| 357 |
+
inputs=[image_input, image_url_input, model_choice, confidence_slider],
|
| 358 |
outputs=[output_image, objects_output, unique_objects_output, properties_output, error_output],
|
| 359 |
)
|
|
|
|
| 360 |
# Clear all inputs and outputs
|
| 361 |
clear_btn.click(
|
| 362 |
+
fn=lambda: [None, "", VALID_MODELS[0], 0.5, None, pd.DataFrame(columns=["Object", "Confidence Score"]), pd.DataFrame(columns=["Unique Object", "Confidence Score"]), pd.DataFrame(), None],
|
| 363 |
inputs=None,
|
| 364 |
+
outputs=[image_input, image_url_input, model_choice, confidence_slider, output_image, objects_output, unique_objects_output, properties_output, error_output],
|
| 365 |
)
|
|
|
|
| 366 |
# JSON tab for API-like output
|
| 367 |
with gr.Tab("π JSON"):
|
| 368 |
with gr.Row():
|
|
|
|
| 372 |
# Model selection dropdown
|
| 373 |
url_model_choice = gr.Dropdown(choices=VALID_MODELS, value=VALID_MODELS[0], label="π Select Model")
|
| 374 |
url_model_info = gr.Markdown(f"**Model Info**: {MODEL_DESCRIPTIONS[VALID_MODELS[0]]}")
|
| 375 |
+
# Confidence threshold slider
|
| 376 |
+
confidence_slider_json = gr.Slider(minimum=0, maximum=1, value=0.5, step=0.01, label="Confidence Threshold")
|
| 377 |
# Image upload input
|
| 378 |
image_input_json = gr.Image(type="pil", label="π· Upload Image")
|
| 379 |
# Image URL input
|
| 380 |
image_url_input_json = gr.Textbox(label="π Image URL", placeholder="https://example.com/image.jpg")
|
| 381 |
# Process button
|
| 382 |
url_submit_btn = gr.Button("π Process", variant="primary")
|
| 383 |
+
url_clear_btn = gr.Button("ποΈ Clear", variant="secondary")
|
| 384 |
# Update model info when model changes
|
| 385 |
url_model_choice.change(
|
| 386 |
fn=lambda model_name: f"**Model Info**: {MODEL_DESCRIPTIONS.get(model_name, 'No description available.')}",
|
| 387 |
inputs=url_model_choice,
|
| 388 |
outputs=url_model_info,
|
| 389 |
)
|
|
|
|
| 390 |
# Right column for JSON output
|
| 391 |
with gr.Column(scale=1):
|
| 392 |
# JSON output display
|
| 393 |
url_output = gr.JSON(label="API Response")
|
|
|
|
| 394 |
# Process image and return JSON when Process button is clicked
|
| 395 |
url_submit_btn.click(
|
| 396 |
+
fn=lambda img, url, model, conf: process_image(img, url, model, for_json=True, confidence_threshold=conf),
|
| 397 |
+
inputs=[image_input_json, image_url_input_json, url_model_choice, confidence_slider_json],
|
| 398 |
outputs=[url_output],
|
| 399 |
)
|
| 400 |
+
# Clear inputs and output
|
| 401 |
+
url_clear_btn.click(
|
| 402 |
+
fn=lambda: [None, "", VALID_MODELS[0], 0.5, {}],
|
| 403 |
+
inputs=None,
|
| 404 |
+
outputs=[image_input_json, image_url_input_json, url_model_choice, confidence_slider_json, url_output],
|
| 405 |
+
)
|
| 406 |
# Help tab with usage instructions
|
| 407 |
with gr.Tab("βΉοΈ Help"):
|
| 408 |
gr.Markdown(
|
| 409 |
"""
|
| 410 |
## How to Use
|
| 411 |
+
- **Home**: Select a model and confidence threshold, upload an image or provide a URL, click "Detect" to see results.
|
| 412 |
+
- **JSON**: Select a model and confidence threshold, upload an image or enter a URL, click "Process" for JSON output.
|
| 413 |
+
- **Models**: Choose DETR (detection or panoptic), YOLOS (lightweight detection), or Deformable DETR (efficient attention).
|
| 414 |
+
- **Clear**: Reset inputs/outputs in Home or JSON tab.
|
| 415 |
- **Errors**: Check error box (Home) or JSON response (JSON) for issues.
|
|
|
|
| 416 |
## Tips
|
| 417 |
- Use high-quality images for better results.
|
| 418 |
- Panoptic models provide segmentation masks for complex scenes.
|
| 419 |
- YOLOS-Tiny is faster for resource-constrained devices.
|
| 420 |
+
- Adjust confidence threshold to filter detections (higher = fewer but more confident).
|
| 421 |
"""
|
| 422 |
)
|
|
|
|
| 423 |
return demo
|
|
|
|
| 424 |
except Exception as e:
|
| 425 |
logger.error(f"Error creating Gradio UI: {str(e)}\n{traceback.format_exc()}")
|
| 426 |
raise RuntimeError(f"Failed to create Gradio UI: {str(e)}")
|
|
|
|
| 427 |
# ------------------------------
|
| 428 |
# Launcher
|
| 429 |
# ------------------------------
|
|
|
|
| 430 |
def parse_args() -> argparse.Namespace:
|
| 431 |
"""
|
| 432 |
Parse command-line arguments for configuring the application.
|
|
|
|
| 433 |
Returns:
|
| 434 |
Parsed arguments as a Namespace object.
|
| 435 |
"""
|
|
|
|
| 448 |
if not 0 <= args.confidence_threshold <= 1:
|
| 449 |
parser.error("Confidence threshold must be between 0 and 1.")
|
| 450 |
return args
|
|
|
|
| 451 |
def find_available_port(start_port: int, port_range: range, max_attempts: int) -> Optional[int]:
|
| 452 |
"""
|
| 453 |
Find an available port within the specified range.
|
|
|
|
| 454 |
Args:
|
| 455 |
start_port: Initial port to try.
|
| 456 |
port_range: Range of ports to attempt.
|
| 457 |
max_attempts: Maximum number of ports to try.
|
|
|
|
| 458 |
Returns:
|
| 459 |
Available port number, or None if no port is found.
|
| 460 |
"""
|
|
|
|
| 478 |
raise
|
| 479 |
logger.error(f"No available port in range {min(port_range)}-{max(port_range)}")
|
| 480 |
return None
|
|
|
|
| 481 |
def main() -> None:
|
| 482 |
"""
|
| 483 |
Launch the Gradio UI and optional FastAPI server.
|
|
|
|
| 484 |
Raises:
|
| 485 |
SystemExit: On interruption or critical errors.
|
| 486 |
"""
|
|
|
|
| 495 |
if gradio_port is None:
|
| 496 |
logger.error("Failed to find an available port for Gradio UI")
|
| 497 |
sys.exit(1)
|
|
|
|
| 498 |
# Start FastAPI server in a thread if enabled
|
| 499 |
if args.enable_fastapi:
|
| 500 |
logger.info(f"Starting FastAPI on port {args.fastapi_port}")
|
|
|
|
| 503 |
daemon=True
|
| 504 |
)
|
| 505 |
fastapi_thread.start()
|
|
|
|
| 506 |
# Launch Gradio UI
|
| 507 |
logger.info(f"Starting Gradio UI on port {gradio_port}")
|
| 508 |
demo = create_gradio_ui()
|
| 509 |
demo.launch(server_port=gradio_port, server_name="0.0.0.0")
|
|
|
|
| 510 |
except KeyboardInterrupt:
|
| 511 |
logger.info("Application terminated by user.")
|
| 512 |
sys.exit(0)
|
| 513 |
except Exception as e:
|
| 514 |
logger.error(f"Error: {str(e)}\n{traceback.format_exc()}")
|
| 515 |
sys.exit(1)
|
|
|
|
| 516 |
if __name__ == "__main__":
|
| 517 |
main()
|