Upload lora-scripts/sd-scripts/finetune/make_captions_by_git.py with huggingface_hub
Browse files
lora-scripts/sd-scripts/finetune/make_captions_by_git.py
ADDED
|
@@ -0,0 +1,183 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import argparse
|
| 2 |
+
import os
|
| 3 |
+
import re
|
| 4 |
+
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
from PIL import Image
|
| 7 |
+
from tqdm import tqdm
|
| 8 |
+
|
| 9 |
+
import torch
|
| 10 |
+
from library.device_utils import init_ipex, get_preferred_device
|
| 11 |
+
init_ipex()
|
| 12 |
+
|
| 13 |
+
from transformers import AutoProcessor, AutoModelForCausalLM
|
| 14 |
+
from transformers.generation.utils import GenerationMixin
|
| 15 |
+
|
| 16 |
+
import library.train_util as train_util
|
| 17 |
+
from library.utils import setup_logging
|
| 18 |
+
setup_logging()
|
| 19 |
+
import logging
|
| 20 |
+
logger = logging.getLogger(__name__)
|
| 21 |
+
|
| 22 |
+
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 23 |
+
|
| 24 |
+
PATTERN_REPLACE = [
|
| 25 |
+
re.compile(r'(has|with|and) the (words?|letters?|name) (" ?[^"]*"|\w+)( ?(is )?(on|in) (the |her |their |him )?\w+)?'),
|
| 26 |
+
re.compile(r'(with a sign )?that says ?(" ?[^"]*"|\w+)( ?on it)?'),
|
| 27 |
+
re.compile(r"(with a sign )?that says ?(' ?(i'm)?[^']*'|\w+)( ?on it)?"),
|
| 28 |
+
re.compile(r"with the number \d+ on (it|\w+ \w+)"),
|
| 29 |
+
re.compile(r'with the words "'),
|
| 30 |
+
re.compile(r"word \w+ on it"),
|
| 31 |
+
re.compile(r"that says the word \w+ on it"),
|
| 32 |
+
re.compile("that says'the word \"( on it)?"),
|
| 33 |
+
]
|
| 34 |
+
|
| 35 |
+
# 誤検知しまくりの with the word xxxx を消す
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
def remove_words(captions, debug):
|
| 39 |
+
removed_caps = []
|
| 40 |
+
for caption in captions:
|
| 41 |
+
cap = caption
|
| 42 |
+
for pat in PATTERN_REPLACE:
|
| 43 |
+
cap = pat.sub("", cap)
|
| 44 |
+
if debug and cap != caption:
|
| 45 |
+
logger.info(caption)
|
| 46 |
+
logger.info(cap)
|
| 47 |
+
removed_caps.append(cap)
|
| 48 |
+
return removed_caps
|
| 49 |
+
|
| 50 |
+
|
| 51 |
+
def collate_fn_remove_corrupted(batch):
|
| 52 |
+
"""Collate function that allows to remove corrupted examples in the
|
| 53 |
+
dataloader. It expects that the dataloader returns 'None' when that occurs.
|
| 54 |
+
The 'None's in the batch are removed.
|
| 55 |
+
"""
|
| 56 |
+
# Filter out all the Nones (corrupted examples)
|
| 57 |
+
batch = list(filter(lambda x: x is not None, batch))
|
| 58 |
+
return batch
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
def main(args):
|
| 62 |
+
r"""
|
| 63 |
+
transformers 4.30.2で、バッチサイズ>1でも動くようになったので、以下コメントアウト
|
| 64 |
+
|
| 65 |
+
# GITにバッチサイズが1より大きくても動くようにパッチを当てる: transformers 4.26.0用
|
| 66 |
+
org_prepare_input_ids_for_generation = GenerationMixin._prepare_input_ids_for_generation
|
| 67 |
+
curr_batch_size = [args.batch_size] # ループの最後で件数がbatch_size未満になるので入れ替えられるように
|
| 68 |
+
|
| 69 |
+
# input_idsがバッチサイズと同じ件数である必要がある:バッチサイズはこの関数から参照できないので外から渡す
|
| 70 |
+
# ここより上で置き換えようとするとすごく大変
|
| 71 |
+
def _prepare_input_ids_for_generation_patch(self, bos_token_id, encoder_outputs):
|
| 72 |
+
input_ids = org_prepare_input_ids_for_generation(self, bos_token_id, encoder_outputs)
|
| 73 |
+
if input_ids.size()[0] != curr_batch_size[0]:
|
| 74 |
+
input_ids = input_ids.repeat(curr_batch_size[0], 1)
|
| 75 |
+
return input_ids
|
| 76 |
+
|
| 77 |
+
GenerationMixin._prepare_input_ids_for_generation = _prepare_input_ids_for_generation_patch
|
| 78 |
+
"""
|
| 79 |
+
|
| 80 |
+
logger.info(f"load images from {args.train_data_dir}")
|
| 81 |
+
train_data_dir_path = Path(args.train_data_dir)
|
| 82 |
+
image_paths = train_util.glob_images_pathlib(train_data_dir_path, args.recursive)
|
| 83 |
+
logger.info(f"found {len(image_paths)} images.")
|
| 84 |
+
|
| 85 |
+
# できればcacheに依存せず明示的にダウンロードしたい
|
| 86 |
+
logger.info(f"loading GIT: {args.model_id}")
|
| 87 |
+
git_processor = AutoProcessor.from_pretrained(args.model_id)
|
| 88 |
+
git_model = AutoModelForCausalLM.from_pretrained(args.model_id).to(DEVICE)
|
| 89 |
+
logger.info("GIT loaded")
|
| 90 |
+
|
| 91 |
+
# captioningする
|
| 92 |
+
def run_batch(path_imgs):
|
| 93 |
+
imgs = [im for _, im in path_imgs]
|
| 94 |
+
|
| 95 |
+
# curr_batch_size[0] = len(path_imgs)
|
| 96 |
+
inputs = git_processor(images=imgs, return_tensors="pt").to(DEVICE) # 画像はpil形式
|
| 97 |
+
generated_ids = git_model.generate(pixel_values=inputs.pixel_values, max_length=args.max_length)
|
| 98 |
+
captions = git_processor.batch_decode(generated_ids, skip_special_tokens=True)
|
| 99 |
+
|
| 100 |
+
if args.remove_words:
|
| 101 |
+
captions = remove_words(captions, args.debug)
|
| 102 |
+
|
| 103 |
+
for (image_path, _), caption in zip(path_imgs, captions):
|
| 104 |
+
with open(os.path.splitext(image_path)[0] + args.caption_extension, "wt", encoding="utf-8") as f:
|
| 105 |
+
f.write(caption + "\n")
|
| 106 |
+
if args.debug:
|
| 107 |
+
logger.info(f"{image_path} {caption}")
|
| 108 |
+
|
| 109 |
+
# 読み込みの高速化のためにDataLoaderを使うオプション
|
| 110 |
+
if args.max_data_loader_n_workers is not None:
|
| 111 |
+
dataset = train_util.ImageLoadingDataset(image_paths)
|
| 112 |
+
data = torch.utils.data.DataLoader(
|
| 113 |
+
dataset,
|
| 114 |
+
batch_size=args.batch_size,
|
| 115 |
+
shuffle=False,
|
| 116 |
+
num_workers=args.max_data_loader_n_workers,
|
| 117 |
+
collate_fn=collate_fn_remove_corrupted,
|
| 118 |
+
drop_last=False,
|
| 119 |
+
)
|
| 120 |
+
else:
|
| 121 |
+
data = [[(None, ip)] for ip in image_paths]
|
| 122 |
+
|
| 123 |
+
b_imgs = []
|
| 124 |
+
for data_entry in tqdm(data, smoothing=0.0):
|
| 125 |
+
for data in data_entry:
|
| 126 |
+
if data is None:
|
| 127 |
+
continue
|
| 128 |
+
|
| 129 |
+
image, image_path = data
|
| 130 |
+
if image is None:
|
| 131 |
+
try:
|
| 132 |
+
image = Image.open(image_path)
|
| 133 |
+
if image.mode != "RGB":
|
| 134 |
+
image = image.convert("RGB")
|
| 135 |
+
except Exception as e:
|
| 136 |
+
logger.error(f"Could not load image path / 画像を読み込めません: {image_path}, error: {e}")
|
| 137 |
+
continue
|
| 138 |
+
|
| 139 |
+
b_imgs.append((image_path, image))
|
| 140 |
+
if len(b_imgs) >= args.batch_size:
|
| 141 |
+
run_batch(b_imgs)
|
| 142 |
+
b_imgs.clear()
|
| 143 |
+
|
| 144 |
+
if len(b_imgs) > 0:
|
| 145 |
+
run_batch(b_imgs)
|
| 146 |
+
|
| 147 |
+
logger.info("done!")
|
| 148 |
+
|
| 149 |
+
|
| 150 |
+
def setup_parser() -> argparse.ArgumentParser:
|
| 151 |
+
parser = argparse.ArgumentParser()
|
| 152 |
+
parser.add_argument("train_data_dir", type=str, help="directory for train images / 学習画像データのディレクトリ")
|
| 153 |
+
parser.add_argument("--caption_extension", type=str, default=".caption", help="extension of caption file / 出力されるキャプションファイルの拡張子")
|
| 154 |
+
parser.add_argument(
|
| 155 |
+
"--model_id",
|
| 156 |
+
type=str,
|
| 157 |
+
default="microsoft/git-large-textcaps",
|
| 158 |
+
help="model id for GIT in Hugging Face / 使用するGITのHugging FaceのモデルID",
|
| 159 |
+
)
|
| 160 |
+
parser.add_argument("--batch_size", type=int, default=1, help="batch size in inference / 推論時のバッチサイズ")
|
| 161 |
+
parser.add_argument(
|
| 162 |
+
"--max_data_loader_n_workers",
|
| 163 |
+
type=int,
|
| 164 |
+
default=None,
|
| 165 |
+
help="enable image reading by DataLoader with this number of workers (faster) / DataLoaderによる画像読み込みを有効にしてこのワーカー数を適用する(読み込みを高速化)",
|
| 166 |
+
)
|
| 167 |
+
parser.add_argument("--max_length", type=int, default=50, help="max length of caption / captionの最大長")
|
| 168 |
+
parser.add_argument(
|
| 169 |
+
"--remove_words",
|
| 170 |
+
action="store_true",
|
| 171 |
+
help="remove like `with the words xxx` from caption / `with the words xxx`のような部分をキャプションから削除する",
|
| 172 |
+
)
|
| 173 |
+
parser.add_argument("--debug", action="store_true", help="debug mode")
|
| 174 |
+
parser.add_argument("--recursive", action="store_true", help="search for images in subfolders recursively / サブフォルダを再帰的に検索する")
|
| 175 |
+
|
| 176 |
+
return parser
|
| 177 |
+
|
| 178 |
+
|
| 179 |
+
if __name__ == "__main__":
|
| 180 |
+
parser = setup_parser()
|
| 181 |
+
|
| 182 |
+
args = parser.parse_args()
|
| 183 |
+
main(args)
|