Spaces:
Sleeping
Sleeping
File size: 14,134 Bytes
d0c8d86 | 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 | """
Caption Model Module
Manages BLIP and GIT models for image caption generation.
Handles model loading, inference, and memory management.
"""
import torch
from PIL import Image
from typing import Optional, Dict, Tuple
from transformers import (
BlipProcessor,
BlipForConditionalGeneration,
AutoProcessor,
AutoModelForCausalLM
)
import gc
from config import model_config
class CaptionModelError(Exception):
"""Custom exception for caption model errors"""
pass
class CaptionModel:
"""
Base class for caption generation models
Provides common interface for BLIP and GIT models
"""
def __init__(self, model_name: str, device: str = "cuda"):
"""
Initialize caption model
Args:
model_name: HuggingFace model identifier
device: Device to load model on (cuda/cpu)
"""
self.model_name = model_name
self.device = self._get_device(device)
self.processor = None
self.model = None
self._is_loaded = False
def _get_device(self, requested_device: str) -> str:
"""
Determine available device
Args:
requested_device: Requested device (cuda/cpu)
Returns:
str: Available device
"""
if requested_device == "cuda" and torch.cuda.is_available():
return "cuda"
return "cpu"
def load(self) -> bool:
"""
Load model into memory
Returns:
bool: True if successful
"""
raise NotImplementedError("Subclass must implement load()")
def generate_caption(
self,
image: Image.Image,
max_length: int = 50,
num_beams: int = 3
) -> str:
"""
Generate caption for image
Args:
image: PIL Image
max_length: Maximum caption length
num_beams: Number of beams for beam search
Returns:
str: Generated caption
"""
raise NotImplementedError("Subclass must implement generate_caption()")
def unload(self) -> None:
"""Unload model from memory"""
if self.model is not None:
del self.model
self.model = None
if self.processor is not None:
del self.processor
self.processor = None
gc.collect()
if self.device == "cuda":
torch.cuda.empty_cache()
self._is_loaded = False
def is_loaded(self) -> bool:
"""Check if model is loaded"""
return self._is_loaded
def get_info(self) -> dict:
"""Get model information"""
return {
"model_name": self.model_name,
"device": self.device,
"is_loaded": self._is_loaded
}
class BLIPModel(CaptionModel):
"""
BLIP (Bootstrapping Language-Image Pre-training) model
Fast and efficient model for image captioning
"""
def __init__(self, device: str = "cuda"):
"""Initialize BLIP model"""
super().__init__(model_config.BLIP_MODEL_NAME, device)
self.max_length = model_config.BLIP_MAX_LENGTH
self.num_beams = model_config.BLIP_NUM_BEAMS
def load(self) -> bool:
"""
Load BLIP model and processor
Returns:
bool: True if successful
"""
try:
print(f"Loading BLIP model on {self.device}...")
# Load processor
self.processor = BlipProcessor.from_pretrained(
self.model_name,
cache_dir=model_config.MODEL_CACHE_DIR
)
# Load model
self.model = BlipForConditionalGeneration.from_pretrained(
self.model_name,
cache_dir=model_config.MODEL_CACHE_DIR,
torch_dtype=torch.float16 if self.device == "cuda" else torch.float32
).to(self.device)
# Set to evaluation mode
self.model.eval()
self._is_loaded = True
print(f"✓ BLIP model loaded successfully on {self.device}")
return True
except Exception as e:
print(f"Error loading BLIP model: {e}")
self._is_loaded = False
return False
def generate_caption(
self,
image: Image.Image,
max_length: Optional[int] = None,
num_beams: Optional[int] = None
) -> str:
"""
Generate caption using BLIP
Args:
image: PIL Image
max_length: Maximum caption length
num_beams: Number of beams for beam search
Returns:
str: Generated caption
Raises:
CaptionModelError: If generation fails
"""
if not self._is_loaded:
raise CaptionModelError("BLIP model not loaded")
try:
# Use default values if not provided
max_length = max_length or self.max_length
num_beams = num_beams or self.num_beams
# Preprocess image
inputs = self.processor(
images=image,
return_tensors="pt"
).to(self.device)
# Generate caption
with torch.no_grad():
output_ids = self.model.generate(
**inputs,
max_length=max_length,
num_beams=num_beams,
early_stopping=True
)
# Decode caption
caption = self.processor.decode(
output_ids[0],
skip_special_tokens=True
)
return caption.strip()
except Exception as e:
raise CaptionModelError(f"BLIP caption generation failed: {e}")
class GITModel(CaptionModel):
"""
GIT (Generative Image-to-text Transformer) model
More detailed and accurate captions compared to BLIP
"""
def __init__(self, device: str = "cuda"):
"""Initialize GIT model"""
super().__init__(model_config.GIT_MODEL_NAME, device)
self.max_length = model_config.GIT_MAX_LENGTH
self.num_beams = model_config.GIT_NUM_BEAMS
def load(self) -> bool:
"""
Load GIT model and processor
Returns:
bool: True if successful
"""
try:
print(f"Loading GIT model on {self.device}...")
# Load processor
self.processor = AutoProcessor.from_pretrained(
self.model_name,
cache_dir=model_config.MODEL_CACHE_DIR
)
# Load model
self.model = AutoModelForCausalLM.from_pretrained(
self.model_name,
cache_dir=model_config.MODEL_CACHE_DIR,
torch_dtype=torch.float16 if self.device == "cuda" else torch.float32
).to(self.device)
# Set to evaluation mode
self.model.eval()
self._is_loaded = True
print(f"✓ GIT model loaded successfully on {self.device}")
return True
except Exception as e:
print(f"Error loading GIT model: {e}")
self._is_loaded = False
return False
def generate_caption(
self,
image: Image.Image,
max_length: Optional[int] = None,
num_beams: Optional[int] = None
) -> str:
"""
Generate caption using GIT
Args:
image: PIL Image
max_length: Maximum caption length
num_beams: Number of beams for beam search
Returns:
str: Generated caption
Raises:
CaptionModelError: If generation fails
"""
if not self._is_loaded:
raise CaptionModelError("GIT model not loaded")
try:
# Use default values if not provided
max_length = max_length or self.max_length
num_beams = num_beams or self.num_beams
# Preprocess image
inputs = self.processor(
images=image,
return_tensors="pt"
).to(self.device)
# Generate caption
with torch.no_grad():
output_ids = self.model.generate(
pixel_values=inputs.pixel_values,
max_length=max_length,
num_beams=num_beams,
early_stopping=True
)
# Decode caption
caption = self.processor.batch_decode(
output_ids,
skip_special_tokens=True
)[0]
return caption.strip()
except Exception as e:
raise CaptionModelError(f"GIT caption generation failed: {e}")
class CaptionModelManager:
"""
Manager for both BLIP and GIT models
Provides unified interface and handles model lifecycle
"""
def __init__(self, device: Optional[str] = None):
"""
Initialize model manager
Args:
device: Device to use (cuda/cpu), auto-detects if None
"""
self.device = device or model_config.DEVICE
# Initialize models
self.blip_model = BLIPModel(self.device)
self.git_model = GITModel(self.device)
# Track which models are loaded
self._loaded_models = set()
def load_all_models(self) -> Tuple[bool, bool]:
"""
Load both models
Returns:
Tuple[bool, bool]: (blip_success, git_success)
"""
blip_success = self.blip_model.load()
if blip_success:
self._loaded_models.add("blip")
git_success = self.git_model.load()
if git_success:
self._loaded_models.add("git")
return blip_success, git_success
def load_model(self, model_name: str) -> bool:
"""
Load specific model
Args:
model_name: Model to load ("blip" or "git")
Returns:
bool: True if successful
"""
if model_name.lower() == "blip":
success = self.blip_model.load()
if success:
self._loaded_models.add("blip")
return success
elif model_name.lower() == "git":
success = self.git_model.load()
if success:
self._loaded_models.add("git")
return success
else:
raise ValueError(f"Unknown model: {model_name}")
def generate_captions(
self,
image: Image.Image
) -> Dict[str, str]:
"""
Generate captions from all loaded models
Args:
image: PIL Image
Returns:
Dict[str, str]: Captions from each model
"""
captions = {}
if "blip" in self._loaded_models:
try:
captions["blip"] = self.blip_model.generate_caption(image)
except Exception as e:
captions["blip"] = f"Error: {str(e)}"
if "git" in self._loaded_models:
try:
captions["git"] = self.git_model.generate_caption(image)
except Exception as e:
captions["git"] = f"Error: {str(e)}"
return captions
def unload_all_models(self) -> None:
"""Unload all models from memory"""
self.blip_model.unload()
self.git_model.unload()
self._loaded_models.clear()
def get_status(self) -> dict:
"""Get status of all models"""
return {
"device": self.device,
"blip": {
"loaded": self.blip_model.is_loaded(),
"info": self.blip_model.get_info()
},
"git": {
"loaded": self.git_model.is_loaded(),
"info": self.git_model.get_info()
},
"loaded_models": list(self._loaded_models)
}
# Singleton instance
_model_manager = None
def get_model_manager() -> CaptionModelManager:
"""Get singleton CaptionModelManager instance"""
global _model_manager
if _model_manager is None:
_model_manager = CaptionModelManager()
return _model_manager
if __name__ == "__main__":
# Test the caption models
print("=" * 60)
print("CAPTION MODELS - TEST MODE")
print("=" * 60)
# Initialize manager
manager = CaptionModelManager()
print(f"\n✓ Model manager initialized")
print(f" Device: {manager.device}")
print("\n" + "=" * 60)
print("Loading models (this may take a few minutes)...")
print("=" * 60)
# Load models
blip_success, git_success = manager.load_all_models()
print(f"\nBLIP: {'✓ Loaded' if blip_success else '✗ Failed'}")
print(f"GIT: {'✓ Loaded' if git_success else '✗ Failed'}")
print("\n" + "=" * 60)
print("Model Status:")
print("=" * 60)
status = manager.get_status()
for key, value in status.items():
if isinstance(value, dict):
print(f"{key}:")
for k, v in value.items():
print(f" {k}: {v}")
else:
print(f"{key}: {value}")
print("\n" + "=" * 60)
print("✓ Caption models test complete")
print("=" * 60)
print("\nTo test caption generation, provide a test image:")
print(" from PIL import Image")
print(" img = Image.open('your_image.jpg')")
print(" captions = manager.generate_captions(img)")
print(" print(captions)") |