Update app.py
Browse files
app.py
CHANGED
|
@@ -1,348 +1,154 @@
|
|
| 1 |
import torch
|
| 2 |
from PIL import Image
|
| 3 |
import torchvision.transforms as T
|
| 4 |
-
from
|
| 5 |
-
from transformers import Blip2Processor, Blip2ForConditionalGeneration
|
| 6 |
import gradio as gr
|
| 7 |
-
import
|
| 8 |
-
from
|
| 9 |
-
import
|
| 10 |
-
from functools import lru_cache
|
| 11 |
|
| 12 |
-
#
|
| 13 |
-
logging.basicConfig(level=logging.INFO)
|
| 14 |
-
logger = logging.getLogger(__name__)
|
| 15 |
-
|
| 16 |
-
# Device setup with memory optimization
|
| 17 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
if device == "cuda":
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
torch.backends.cuda.enable_math_sdp(True)
|
| 22 |
|
| 23 |
-
#
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard',
|
| 29 |
-
'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich',
|
| 30 |
-
'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed',
|
| 31 |
-
'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 'microwave', 'oven',
|
| 32 |
-
'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush'
|
| 33 |
-
]
|
| 34 |
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
self.model = None
|
| 41 |
-
self._load_models()
|
| 42 |
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
# Use appropriate dtype based on device
|
| 61 |
-
dtype = torch.float16 if self.device == "cuda" else torch.float32
|
| 62 |
-
|
| 63 |
-
self.processor = Blip2Processor.from_pretrained(
|
| 64 |
-
model_name,
|
| 65 |
-
torch_dtype=dtype,
|
| 66 |
-
low_cpu_mem_usage=True
|
| 67 |
-
)
|
| 68 |
-
|
| 69 |
-
self.model = Blip2ForConditionalGeneration.from_pretrained(
|
| 70 |
-
model_name,
|
| 71 |
-
torch_dtype=dtype,
|
| 72 |
-
low_cpu_mem_usage=True,
|
| 73 |
-
device_map="auto" if self.device == "cuda" else None
|
| 74 |
)
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
# Enable gradient checkpointing for memory efficiency
|
| 82 |
-
if hasattr(self.model, 'gradient_checkpointing_enable'):
|
| 83 |
-
self.model.gradient_checkpointing_enable()
|
| 84 |
-
|
| 85 |
-
logger.info("Models loaded successfully!")
|
| 86 |
-
|
| 87 |
-
except Exception as e:
|
| 88 |
-
logger.error(f"Error loading models: {e}")
|
| 89 |
-
raise
|
| 90 |
-
|
| 91 |
-
@lru_cache(maxsize=32)
|
| 92 |
-
def _get_transform(self, size: int = 384):
|
| 93 |
-
"""Cached transform to avoid recreation"""
|
| 94 |
-
return T.Compose([
|
| 95 |
-
T.Resize((size, size), interpolation=T.InterpolationMode.BILINEAR),
|
| 96 |
-
T.ToTensor(),
|
| 97 |
-
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
| 98 |
-
])
|
| 99 |
-
|
| 100 |
-
def detect_objects(self, image_tensor: torch.Tensor, confidence_threshold: float = 0.7) -> List[str]:
|
| 101 |
-
"""Optimized object detection with better filtering"""
|
| 102 |
-
try:
|
| 103 |
-
with torch.no_grad():
|
| 104 |
-
predictions = self.detector(image_tensor)[0]
|
| 105 |
-
|
| 106 |
-
# Filter by confidence and remove duplicates
|
| 107 |
-
detected_objects = []
|
| 108 |
-
seen_labels = set()
|
| 109 |
-
|
| 110 |
-
for label_idx, score in zip(predictions["labels"].cpu().tolist(),
|
| 111 |
-
predictions["scores"].cpu().tolist()):
|
| 112 |
-
if score > confidence_threshold and label_idx < len(COCO_LABELS):
|
| 113 |
-
label = COCO_LABELS[label_idx]
|
| 114 |
-
if label != '__background__' and label not in seen_labels:
|
| 115 |
-
detected_objects.append((label, score))
|
| 116 |
-
seen_labels.add(label)
|
| 117 |
-
|
| 118 |
-
# Sort by confidence and return top objects
|
| 119 |
-
detected_objects.sort(key=lambda x: x[1], reverse=True)
|
| 120 |
-
return [obj[0] for obj in detected_objects[:10]] # Limit to top 10
|
| 121 |
-
|
| 122 |
-
except Exception as e:
|
| 123 |
-
logger.warning(f"Object detection failed: {e}")
|
| 124 |
-
return []
|
| 125 |
-
|
| 126 |
-
def generate_caption(self, image: Image.Image, detected_objects: List[str],
|
| 127 |
-
max_new_tokens: int = 32) -> str:
|
| 128 |
-
"""Generate caption with improved prompting"""
|
| 129 |
-
try:
|
| 130 |
-
# Create context-aware prompt
|
| 131 |
-
if detected_objects:
|
| 132 |
-
objects_str = ", ".join(detected_objects[:5]) # Limit for token efficiency
|
| 133 |
-
prompt = f"Objects detected: {objects_str}. Describe this image with specific details about colors, materials, actions, and setting:"
|
| 134 |
-
else:
|
| 135 |
-
prompt = "Describe this image with specific details about what you see, including colors, materials, actions, and setting:"
|
| 136 |
-
|
| 137 |
-
# Process inputs
|
| 138 |
-
inputs = self.processor(
|
| 139 |
-
images=image,
|
| 140 |
-
text=prompt,
|
| 141 |
-
return_tensors="pt",
|
| 142 |
-
padding=True,
|
| 143 |
-
truncation=True,
|
| 144 |
-
max_length=512
|
| 145 |
-
).to(self.device)
|
| 146 |
-
|
| 147 |
-
# Generate with optimized parameters
|
| 148 |
-
with torch.no_grad():
|
| 149 |
-
outputs = self.model.generate(
|
| 150 |
-
**inputs,
|
| 151 |
-
max_new_tokens=max_new_tokens,
|
| 152 |
-
num_beams=3,
|
| 153 |
-
do_sample=False, # Deterministic for consistency
|
| 154 |
-
no_repeat_ngram_size=3,
|
| 155 |
-
early_stopping=True,
|
| 156 |
-
pad_token_id=self.processor.tokenizer.pad_token_id,
|
| 157 |
-
eos_token_id=self.processor.tokenizer.eos_token_id,
|
| 158 |
-
length_penalty=1.0,
|
| 159 |
-
repetition_penalty=1.1
|
| 160 |
-
)
|
| 161 |
-
|
| 162 |
-
# Decode and clean up
|
| 163 |
-
caption = self.processor.decode(outputs[0], skip_special_tokens=True)
|
| 164 |
-
|
| 165 |
-
# Remove the prompt from the output if it's included
|
| 166 |
-
if prompt in caption:
|
| 167 |
-
caption = caption.replace(prompt, "").strip()
|
| 168 |
-
|
| 169 |
-
return caption.strip() if caption.strip() else "Unable to generate description."
|
| 170 |
-
|
| 171 |
-
except Exception as e:
|
| 172 |
-
logger.error(f"Caption generation failed: {e}")
|
| 173 |
-
return f"Error generating caption: {str(e)}"
|
| 174 |
-
|
| 175 |
-
def process_image(self, image_path: str, max_new_tokens: int = 32) -> str:
|
| 176 |
-
"""Main processing pipeline with error handling"""
|
| 177 |
-
try:
|
| 178 |
-
# Load and validate image
|
| 179 |
-
image = Image.open(image_path).convert("RGB")
|
| 180 |
-
|
| 181 |
-
# Check image size and resize if too large
|
| 182 |
-
max_size = 1024
|
| 183 |
-
if max(image.size) > max_size:
|
| 184 |
-
image.thumbnail((max_size, max_size), Image.Resampling.LANCZOS)
|
| 185 |
-
|
| 186 |
-
# Prepare image for object detection
|
| 187 |
-
transform = self._get_transform()
|
| 188 |
-
image_tensor = transform(image).unsqueeze(0).to(self.device)
|
| 189 |
-
|
| 190 |
-
# Detect objects
|
| 191 |
-
detected_objects = self.detect_objects(image_tensor)
|
| 192 |
-
|
| 193 |
-
# Generate caption
|
| 194 |
-
caption = self.generate_caption(image, detected_objects, max_new_tokens)
|
| 195 |
-
|
| 196 |
-
# Clean up GPU memory
|
| 197 |
-
if self.device == "cuda":
|
| 198 |
-
del image_tensor
|
| 199 |
-
torch.cuda.empty_cache()
|
| 200 |
-
|
| 201 |
-
return caption.capitalize() if caption else "No description available."
|
| 202 |
-
|
| 203 |
-
except Exception as e:
|
| 204 |
-
logger.error(f"Error processing image {image_path}: {e}")
|
| 205 |
-
return f"Error processing image: {str(e)}"
|
| 206 |
-
|
| 207 |
-
# Global captioner instance
|
| 208 |
-
captioner = OptimizedImageCaptioner()
|
| 209 |
|
| 210 |
-
def
|
| 211 |
-
"""Process multiple images
|
| 212 |
if not files:
|
| 213 |
return "No images uploaded."
|
| 214 |
|
|
|
|
| 215 |
results = []
|
| 216 |
-
total_files = len(files)
|
| 217 |
|
|
|
|
| 218 |
for i, file in enumerate(files):
|
| 219 |
-
|
| 220 |
-
|
| 221 |
-
|
| 222 |
-
|
| 223 |
-
|
| 224 |
-
|
| 225 |
-
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
except Exception as e:
|
| 230 |
-
logger.error(f"Failed to process {file.name}: {e}")
|
| 231 |
-
results.append(f"Image {i+1}: Error - {str(e)}")
|
| 232 |
|
|
|
|
| 233 |
return "\n\n".join(results)
|
| 234 |
|
| 235 |
-
def
|
| 236 |
-
"""
|
| 237 |
if not files:
|
| 238 |
return []
|
| 239 |
|
| 240 |
previews = []
|
| 241 |
-
for file in files[:
|
| 242 |
try:
|
| 243 |
img = Image.open(file.name).convert("RGB")
|
| 244 |
-
|
| 245 |
-
img.thumbnail((300, 300), Image.Resampling.LANCZOS)
|
| 246 |
previews.append(img)
|
| 247 |
-
except
|
| 248 |
-
|
| 249 |
|
| 250 |
return previews
|
| 251 |
|
| 252 |
-
# Gradio Interface
|
| 253 |
with gr.Blocks(
|
| 254 |
-
title="
|
| 255 |
-
theme=gr.themes.
|
| 256 |
-
css="""
|
| 257 |
-
.gradio-container {
|
| 258 |
-
max-width: 1200px !important;
|
| 259 |
-
}
|
| 260 |
-
"""
|
| 261 |
) as demo:
|
| 262 |
|
| 263 |
gr.Markdown("""
|
| 264 |
-
#
|
| 265 |
|
| 266 |
-
|
| 267 |
-
- **Object Detection**: Identifies objects in images for better context
|
| 268 |
-
- **Advanced Captioning**: Uses BLIP-2 for detailed, natural descriptions
|
| 269 |
-
- **Batch Processing**: Handle multiple images at once
|
| 270 |
""")
|
| 271 |
|
| 272 |
with gr.Row():
|
| 273 |
-
with gr.Column(
|
| 274 |
file_input = gr.File(
|
| 275 |
-
label="
|
| 276 |
file_count="multiple",
|
| 277 |
-
file_types=["image"]
|
| 278 |
-
height=200
|
| 279 |
)
|
| 280 |
|
| 281 |
-
|
| 282 |
-
|
| 283 |
-
|
| 284 |
-
|
| 285 |
-
|
| 286 |
-
|
| 287 |
-
|
| 288 |
-
info="Higher = more detailed descriptions"
|
| 289 |
-
)
|
| 290 |
|
| 291 |
with gr.Row():
|
| 292 |
-
generate_btn = gr.Button(
|
| 293 |
-
|
| 294 |
-
variant="primary",
|
| 295 |
-
size="lg"
|
| 296 |
-
)
|
| 297 |
-
clear_btn = gr.Button(
|
| 298 |
-
"��️ Clear All",
|
| 299 |
-
variant="secondary"
|
| 300 |
-
)
|
| 301 |
|
| 302 |
-
with gr.Column(
|
| 303 |
gallery = gr.Gallery(
|
| 304 |
-
label="
|
| 305 |
columns=3,
|
| 306 |
-
|
| 307 |
-
height=400,
|
| 308 |
-
object_fit="contain"
|
| 309 |
)
|
| 310 |
|
| 311 |
-
|
| 312 |
-
label="
|
| 313 |
-
lines=
|
| 314 |
-
|
| 315 |
-
show_copy_button=True,
|
| 316 |
-
placeholder="Descriptions will appear here after processing..."
|
| 317 |
-
)
|
| 318 |
-
|
| 319 |
-
# Event handlers
|
| 320 |
-
file_input.change(
|
| 321 |
-
fn=preview_images,
|
| 322 |
-
inputs=file_input,
|
| 323 |
-
outputs=gallery,
|
| 324 |
-
show_progress=True
|
| 325 |
)
|
| 326 |
|
| 327 |
-
|
| 328 |
-
|
| 329 |
-
|
| 330 |
-
|
| 331 |
-
show_progress=True
|
| 332 |
-
)
|
| 333 |
-
|
| 334 |
-
clear_btn.click(
|
| 335 |
-
fn=lambda: (None, [], 48, ""),
|
| 336 |
-
outputs=[file_input, gallery, max_tokens_slider, output_textbox]
|
| 337 |
-
)
|
| 338 |
|
| 339 |
-
# Launch configuration
|
| 340 |
if __name__ == "__main__":
|
| 341 |
-
demo.launch(
|
| 342 |
-
server_name="0.0.0.0",
|
| 343 |
-
server_port=7860,
|
| 344 |
-
share=False,
|
| 345 |
-
debug=False,
|
| 346 |
-
show_error=True,
|
| 347 |
-
inbrowser=True
|
| 348 |
-
)
|
|
|
|
| 1 |
import torch
|
| 2 |
from PIL import Image
|
| 3 |
import torchvision.transforms as T
|
| 4 |
+
from transformers import BlipProcessor, BlipForConditionalGeneration
|
|
|
|
| 5 |
import gradio as gr
|
| 6 |
+
import asyncio
|
| 7 |
+
from concurrent.futures import ThreadPoolExecutor
|
| 8 |
+
import time
|
|
|
|
| 9 |
|
| 10 |
+
# Device setup
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 12 |
+
print(f"Using device: {device}")
|
| 13 |
+
|
| 14 |
+
# Use BLIP-1 instead of BLIP-2 for much faster inference
|
| 15 |
+
model_name = "Salesforce/blip-image-captioning-base"
|
| 16 |
+
processor = BlipProcessor.from_pretrained(model_name)
|
| 17 |
+
model = BlipForConditionalGeneration.from_pretrained(model_name)
|
| 18 |
+
|
| 19 |
+
# Move to device and optimize
|
| 20 |
+
model.to(device)
|
| 21 |
+
model.eval()
|
| 22 |
+
|
| 23 |
+
# Enable inference optimizations
|
| 24 |
if device == "cuda":
|
| 25 |
+
model = model.half() # Use half precision for speed
|
| 26 |
+
torch.backends.cudnn.benchmark = True
|
|
|
|
| 27 |
|
| 28 |
+
# Fast preprocessing - smaller size for speed
|
| 29 |
+
transform = T.Compose([
|
| 30 |
+
T.Resize((224, 224)), # Smaller size = faster processing
|
| 31 |
+
T.ToTensor(),
|
| 32 |
+
])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
|
| 34 |
+
def generate_fast_caption(image_path, max_length=20):
|
| 35 |
+
"""Ultra-fast caption generation"""
|
| 36 |
+
try:
|
| 37 |
+
# Load image
|
| 38 |
+
image = Image.open(image_path).convert("RGB")
|
|
|
|
|
|
|
| 39 |
|
| 40 |
+
# Resize if too large (speed optimization)
|
| 41 |
+
if max(image.size) > 512:
|
| 42 |
+
image.thumbnail((512, 512), Image.Resampling.LANCZOS)
|
| 43 |
+
|
| 44 |
+
# Simple prompt for basic captioning
|
| 45 |
+
inputs = processor(image, return_tensors="pt").to(device)
|
| 46 |
+
|
| 47 |
+
# Fast generation with minimal beams
|
| 48 |
+
with torch.no_grad():
|
| 49 |
+
outputs = model.generate(
|
| 50 |
+
**inputs,
|
| 51 |
+
max_length=max_length,
|
| 52 |
+
num_beams=1, # Single beam for speed
|
| 53 |
+
do_sample=False,
|
| 54 |
+
early_stopping=True,
|
| 55 |
+
pad_token_id=processor.tokenizer.pad_token_id
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
)
|
| 57 |
+
|
| 58 |
+
caption = processor.decode(outputs[0], skip_special_tokens=True)
|
| 59 |
+
return caption.strip()
|
| 60 |
+
|
| 61 |
+
except Exception as e:
|
| 62 |
+
return f"Error: {str(e)}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
|
| 64 |
+
def process_images_fast(files, max_length):
|
| 65 |
+
"""Process multiple images quickly"""
|
| 66 |
if not files:
|
| 67 |
return "No images uploaded."
|
| 68 |
|
| 69 |
+
start_time = time.time()
|
| 70 |
results = []
|
|
|
|
| 71 |
|
| 72 |
+
# Process images sequentially for stability
|
| 73 |
for i, file in enumerate(files):
|
| 74 |
+
caption = generate_fast_caption(file.name, max_length=int(max_length))
|
| 75 |
+
results.append(f"Image {i+1}: {caption}")
|
| 76 |
+
|
| 77 |
+
# Quick memory cleanup for CUDA
|
| 78 |
+
if device == "cuda" and i % 5 == 0:
|
| 79 |
+
torch.cuda.empty_cache()
|
| 80 |
+
|
| 81 |
+
end_time = time.time()
|
| 82 |
+
processing_time = end_time - start_time
|
|
|
|
|
|
|
|
|
|
|
|
|
| 83 |
|
| 84 |
+
results.append(f"\n⚡ Processed {len(files)} images in {processing_time:.2f} seconds")
|
| 85 |
return "\n\n".join(results)
|
| 86 |
|
| 87 |
+
def preview_images_fast(files):
|
| 88 |
+
"""Quick image preview"""
|
| 89 |
if not files:
|
| 90 |
return []
|
| 91 |
|
| 92 |
previews = []
|
| 93 |
+
for file in files[:12]: # Limit for speed
|
| 94 |
try:
|
| 95 |
img = Image.open(file.name).convert("RGB")
|
| 96 |
+
img.thumbnail((200, 200), Image.Resampling.LANCZOS)
|
|
|
|
| 97 |
previews.append(img)
|
| 98 |
+
except:
|
| 99 |
+
continue
|
| 100 |
|
| 101 |
return previews
|
| 102 |
|
| 103 |
+
# Streamlined Gradio Interface
|
| 104 |
with gr.Blocks(
|
| 105 |
+
title="⚡ Fast Image Captioning",
|
| 106 |
+
theme=gr.themes.Default()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 107 |
) as demo:
|
| 108 |
|
| 109 |
gr.Markdown("""
|
| 110 |
+
# ⚡ Ultra-Fast Image Captioning
|
| 111 |
|
| 112 |
+
Quick and simple image descriptions optimized for speed!
|
|
|
|
|
|
|
|
|
|
| 113 |
""")
|
| 114 |
|
| 115 |
with gr.Row():
|
| 116 |
+
with gr.Column():
|
| 117 |
file_input = gr.File(
|
| 118 |
+
label="Upload Images",
|
| 119 |
file_count="multiple",
|
| 120 |
+
file_types=["image"]
|
|
|
|
| 121 |
)
|
| 122 |
|
| 123 |
+
max_length_slider = gr.Slider(
|
| 124 |
+
minimum=10,
|
| 125 |
+
maximum=30,
|
| 126 |
+
step=5,
|
| 127 |
+
value=20,
|
| 128 |
+
label="Caption Length"
|
| 129 |
+
)
|
|
|
|
|
|
|
| 130 |
|
| 131 |
with gr.Row():
|
| 132 |
+
generate_btn = gr.Button("🚀 Generate Fast", variant="primary")
|
| 133 |
+
clear_btn = gr.Button("Clear")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 134 |
|
| 135 |
+
with gr.Column():
|
| 136 |
gallery = gr.Gallery(
|
| 137 |
+
label="Preview",
|
| 138 |
columns=3,
|
| 139 |
+
height=300
|
|
|
|
|
|
|
| 140 |
)
|
| 141 |
|
| 142 |
+
output = gr.Textbox(
|
| 143 |
+
label="Captions",
|
| 144 |
+
lines=6,
|
| 145 |
+
show_copy_button=True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
)
|
| 147 |
|
| 148 |
+
# Events
|
| 149 |
+
file_input.change(preview_images_fast, file_input, gallery)
|
| 150 |
+
generate_btn.click(process_images_fast, [file_input, max_length_slider], output)
|
| 151 |
+
clear_btn.click(lambda: (None, [], ""), outputs=[file_input, gallery, output])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 152 |
|
|
|
|
| 153 |
if __name__ == "__main__":
|
| 154 |
+
demo.launch(share=False, inbrowser=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|