File size: 12,134 Bytes
fb83c5b |
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 |
from PIL import Image
from transformers import Blip2Processor, Blip2ForConditionalGeneration
import torch
import gradio as gr
import os
from .common_gui import get_folder_path, scriptdir, list_dirs
from .custom_logging import setup_logging
# Set up logging
log = setup_logging()
def load_model():
# Set the device to GPU if available, otherwise use CPU
device = "cuda" if torch.cuda.is_available() else "cpu"
# Initialize the BLIP2 processor
processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
# Initialize the BLIP2 model
model = Blip2ForConditionalGeneration.from_pretrained(
"Salesforce/blip2-opt-2.7b", torch_dtype=torch.float16
)
# Move the model to the specified device
model.to(device)
return processor, model, device
def get_images_in_directory(directory_path):
"""
Returns a list of image file paths found in the provided directory path.
Parameters:
- directory_path: A string representing the path to the directory to search for images.
Returns:
- A list of strings, where each string is the full path to an image file found in the specified directory.
"""
import os
# List of common image file extensions to look for
image_extensions = [".jpg", ".jpeg", ".png", ".bmp", ".gif"]
# Generate a list of image file paths in the directory
image_files = [
# constructs the full path to the file
os.path.join(directory_path, file)
# lists all files and directories in the given path
for file in os.listdir(directory_path)
# gets the file extension in lowercase
if os.path.splitext(file)[1].lower() in image_extensions
]
# Return the list of image file paths
return image_files
def generate_caption(
file_list,
processor,
model,
device,
caption_file_ext=".txt",
num_beams=5,
repetition_penalty=1.5,
length_penalty=1.2,
max_new_tokens=40,
min_new_tokens=20,
do_sample=True,
temperature=1.0,
top_p=0.0,
):
"""
Fetches and processes each image in file_list, generates captions based on the image, and writes the generated captions to a file.
Parameters:
- file_list: A list of file paths pointing to the images to be captioned.
- processor: The preprocessor for the BLIP2 model.
- model: The BLIP2 model to be used for generating captions.
- device: The device on which the computation is performed.
- extension: The extension for the output text files.
- num_beams: Number of beams for beam search. Default: 5.
- repetition_penalty: Penalty for repeating tokens. Default: 1.5.
- length_penalty: Penalty for sentence length. Default: 1.2.
- max_new_tokens: Maximum number of new tokens to generate. Default: 40.
- min_new_tokens: Minimum number of new tokens to generate. Default: 20.
"""
for file_path in file_list:
image = Image.open(file_path)
inputs = processor(images=image, return_tensors="pt").to(device, torch.float16)
if top_p == 0.0:
generated_ids = model.generate(
**inputs,
num_beams=num_beams,
repetition_penalty=repetition_penalty,
length_penalty=length_penalty,
max_new_tokens=max_new_tokens,
min_new_tokens=min_new_tokens,
)
else:
generated_ids = model.generate(
**inputs,
do_sample=do_sample,
top_p=top_p,
max_new_tokens=max_new_tokens,
min_new_tokens=min_new_tokens,
temperature=temperature,
)
generated_text = processor.batch_decode(
generated_ids, skip_special_tokens=True
)[0].strip()
# Construct the output file path by replacing the original file extension with the specified extension
output_file_path = os.path.splitext(file_path)[0] + caption_file_ext
# Write the generated text to the output file
with open(output_file_path, "w", encoding="utf-8") as output_file:
output_file.write(generated_text)
# Log the image file path with a message about the fact that the caption was generated
log.info(f"{file_path} caption was generated")
def caption_images_beam_search(
directory_path,
num_beams,
repetition_penalty,
length_penalty,
min_new_tokens,
max_new_tokens,
caption_file_ext,
):
"""
Captions all images in the specified directory using the provided prompt.
Parameters:
- directory_path: A string representing the path to the directory containing the images to be captioned.
"""
log.info("BLIP2 captionning beam...")
if not os.path.isdir(directory_path):
log.error(f"Directory {directory_path} does not exist.")
return
processor, model, device = load_model()
image_files = get_images_in_directory(directory_path)
generate_caption(
file_list=image_files,
processor=processor,
model=model,
device=device,
num_beams=int(num_beams),
repetition_penalty=float(repetition_penalty),
length_penalty=length_penalty,
min_new_tokens=int(min_new_tokens),
max_new_tokens=int(max_new_tokens),
caption_file_ext=caption_file_ext,
)
def caption_images_nucleus(
directory_path,
do_sample,
temperature,
top_p,
min_new_tokens,
max_new_tokens,
caption_file_ext,
):
"""
Captions all images in the specified directory using the provided prompt.
Parameters:
- directory_path: A string representing the path to the directory containing the images to be captioned.
"""
log.info("BLIP2 captionning nucleus...")
if not os.path.isdir(directory_path):
log.error(f"Directory {directory_path} does not exist.")
return
processor, model, device = load_model()
image_files = get_images_in_directory(directory_path)
generate_caption(
file_list=image_files,
processor=processor,
model=model,
device=device,
do_sample=do_sample,
temperature=temperature,
top_p=top_p,
min_new_tokens=int(min_new_tokens),
max_new_tokens=int(max_new_tokens),
caption_file_ext=caption_file_ext,
)
def gradio_blip2_caption_gui_tab(headless=False, directory_path=None):
from .common_gui import create_refresh_button
directory_path = (
directory_path
if directory_path is not None
else os.path.join(scriptdir, "data")
)
current_train_dir = directory_path
def list_train_dirs(path):
nonlocal current_train_dir
current_train_dir = path
return list(list_dirs(path))
with gr.Tab("BLIP2 Captioning"):
gr.Markdown(
"This utility uses BLIP2 to caption files for each image in a folder."
)
with gr.Group(), gr.Row():
directory_path_dir = gr.Dropdown(
label="Image folder to caption (containing the images to caption)",
choices=[""] + list_train_dirs(directory_path),
value="",
interactive=True,
allow_custom_value=True,
)
create_refresh_button(
directory_path_dir,
lambda: None,
lambda: {"choices": list_train_dirs(current_train_dir)},
"open_folder_small",
)
button_directory_path_dir_input = gr.Button(
"📂",
elem_id="open_folder_small",
elem_classes=["tool"],
visible=(not headless),
)
button_directory_path_dir_input.click(
get_folder_path,
outputs=directory_path_dir,
show_progress=False,
)
with gr.Group(), gr.Row():
min_new_tokens = gr.Number(
value=20,
label="Min new tokens",
interactive=True,
step=1,
minimum=5,
maximum=300,
)
max_new_tokens = gr.Number(
value=40,
label="Max new tokens",
interactive=True,
step=1,
minimum=5,
maximum=300,
)
caption_file_ext = gr.Textbox(
label="Caption file extension",
placeholder="Extension for caption file (e.g., .caption, .txt)",
value=".txt",
interactive=True,
)
with gr.Row():
with gr.Tab("Beam search"):
with gr.Row():
num_beams = gr.Slider(
minimum=1,
maximum=16,
value=16,
step=1,
interactive=True,
label="Number of beams",
)
len_penalty = gr.Slider(
minimum=-1.0,
maximum=2.0,
value=1.0,
step=0.2,
interactive=True,
label="Length Penalty",
info="increase for longer sequence",
)
rep_penalty = gr.Slider(
minimum=1.0,
maximum=5.0,
value=1.5,
step=0.5,
interactive=True,
label="Repeat Penalty",
info="larger value prevents repetition",
)
caption_button_beam = gr.Button(
value="Caption images", interactive=True, variant="primary"
)
caption_button_beam.click(
caption_images_beam_search,
inputs=[
directory_path_dir,
num_beams,
rep_penalty,
len_penalty,
min_new_tokens,
max_new_tokens,
caption_file_ext,
],
)
with gr.Tab("Nucleus sampling"):
with gr.Row():
do_sample = gr.Checkbox(label="Sample", value=True)
temperature = gr.Slider(
minimum=0.5,
maximum=1.0,
value=1.0,
step=0.1,
interactive=True,
label="Temperature",
info="used with nucleus sampling",
)
top_p = gr.Slider(
minimum=0,
maximum=1,
value=0.9,
step=0.1,
interactive=True,
label="Top_p",
)
caption_button_nucleus = gr.Button(
value="Caption images", interactive=True, variant="primary"
)
caption_button_nucleus.click(
caption_images_nucleus,
inputs=[
directory_path_dir,
do_sample,
temperature,
top_p,
min_new_tokens,
max_new_tokens,
caption_file_ext,
],
)
|