dataxl / train /run2.py
Cupret's picture
Upload run2.py
39cd840 verified
Raw
History Blame Contribute Delete
4.08 kB
from typing import Generator, Iterable
from tagger.interrogator import Interrogator
from PIL import Image
from pathlib import Path
import argparse
from tagger.interrogators import interrogators
import os
import shutil
parser = argparse.ArgumentParser()
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument('--dir', help='Predictions for all images in the directory')
group.add_argument('--file', help='Predictions for one file')
parser.add_argument(
'--threshold',
type=float,
default=0.35,
help='Prediction threshold (default is 0.35)')
parser.add_argument(
'--ext',
default='.txt',
help='Extension to add to caption file in case of dir option (default is .txt)')
parser.add_argument(
'--overwrite',
action='store_true',
help='Overwrite caption file if it exists')
parser.add_argument(
'--cpu',
action='store_true',
help='Use CPU only')
parser.add_argument(
'--rawtag',
action='store_true',
help='Use the raw output of the model')
parser.add_argument(
'--recursive',
action='store_true',
help='Enable recursive file search')
parser.add_argument(
'--exclude-tag',
dest='exclude_tags',
action='append',
metavar='t1,t2,t3',
help='Specify tags to exclude (Need comma-separated list)')
parser.add_argument(
'--model',
default='wd14-convnextv2.v1',
choices=list(interrogators.keys()),
help='modelname to use for prediction (default is wd14-convnextv2.v1)')
parser.add_argument(
'--copyto',
default = None,
help='copy_txt')
args = parser.parse_args()
# get interrogator configs
interrogator = interrogators[args.model]
if args.cpu:
interrogator.use_cpu()
def parse_exclude_tags() -> set[str]:
if args.exclude_tags is None:
return set()
tags = []
for str in args.exclude_tags:
for tag in str.split(','):
tags.append(tag.strip())
# reverse escape (nai tag to danbooru tag)
reverse_escaped_tags = []
for tag in tags:
tag = tag.replace(' ', '_').replace('\(', '(').replace('\)', ')')
reverse_escaped_tags.append(tag)
return set([*tags, *reverse_escaped_tags]) # reduce duplicates
def image_interrogate(image_path: Path, tag_escape: bool, exclude_tags: Iterable[str]) -> dict[str, float]:
"""
Predictions from a image path
"""
im = Image.open(image_path)
result = interrogator.interrogate(im)
return Interrogator.postprocess_tags(
result[1],
threshold=args.threshold,
escape_tag=tag_escape,
replace_underscore=tag_escape,
exclude_tags=exclude_tags)
def explore_image_files(folder_path: Path) -> Generator[Path, None, None]:
"""
Explore files by folder path
"""
for path in folder_path.iterdir():
if path.is_file() and path.suffix in ['.png', '.jpg', '.jpeg', '.webp']:
yield path
elif args.recursive and path.is_dir():
yield from explore_image_files(path)
if args.dir:
root_path = Path(args.dir)
for image_path in explore_image_files(root_path):
caption_path = image_path.parent / f'{image_path.stem}{args.ext}'
if caption_path.is_file() and not args.overwrite:
# skip if caption exists
print('skip:', image_path)
continue
print('processing:', image_path)
tags = image_interrogate(image_path, not args.rawtag, parse_exclude_tags())
tags_str = ', '.join(tags.keys())
with open(caption_path, 'w') as fp:
fp.write(tags_str)
if args.copyto:
if(not os.path.exists(args.copyto)):
os.makedirs(args.copyto)
source_path = str(caption_path)
copy_path = os.path.join(args.copyto, source_path.rsplit('/')[-1])
shutil.copy(source_path, copy_path)
print('copy to:', copy_path)
if args.file:
tags = image_interrogate(Path(args.file), not args.rawtag, parse_exclude_tags())
print()
tags_str = ', '.join(tags.keys())
print(tags_str)