Spaces:
Sleeping
Sleeping
Download blockdiffusion/data/filename_captions.py from wkea/blockdiffusion-api: direct link, hf CLI and curl.
- Browser
- Download file 2.24 kB
-
https://huggingface.co/spaces/wkea/blockdiffusion-api/resolve/main/blockdiffusion/data/filename_captions.py
- Command line
-
hf download hf://spaces/wkea/blockdiffusion-api/blockdiffusion/data/filename_captions.py
-
curl -L -o filename_captions.py https://huggingface.co/spaces/wkea/blockdiffusion-api/resolve/main/blockdiffusion/data/filename_captions.py
2.24 kB
| from __future__ import annotations | |
| import os | |
| from typing import List, Optional | |
| from blockdiffusion.data.captions import CaptionItem | |
| _IMG_EXTS = (".png", ".jpg", ".jpeg", ".webp", ".bmp") | |
| def _caption_from_filename(filename: str) -> str: | |
| """ | |
| 从文件名提取“弱标注提示词”。 | |
| 适配命名格式: | |
| - HASH-tag_tag_tag.png | |
| - HASH-side.png / HASH-top.png / HASH-bottom.png | |
| 提取规则: | |
| - 去掉扩展名 | |
| - 以第一个 '-' 作为分隔,取其后部分(若不存在 '-' 则使用全名) | |
| - '_' / '-' 统一当作空格 | |
| """ | |
| name = os.path.splitext(os.path.basename(filename))[0] | |
| if "-" in name: | |
| # 只切一次,避免 tag 本身含 '-' 时被过度切分 | |
| _, tail = name.split("-", 1) | |
| else: | |
| tail = name | |
| tail = tail.replace("_", " ").replace("-", " ").strip().lower() | |
| # 多空格归一 | |
| tail = " ".join([t for t in tail.split(" ") if t]) | |
| return tail | |
| def build_caption_items_from_filenames( | |
| root_dir: str, | |
| verbose: bool = False, | |
| log_every: int = 10_000, | |
| ) -> List[CaptionItem]: | |
| """ | |
| 扫描目录下所有图片(含子目录),用文件名自动生成 captions。 | |
| 返回 CaptionItem 列表,可直接用于 TextureCaptionDataset。 | |
| """ | |
| root_dir = os.path.abspath(root_dir) | |
| items: List[CaptionItem] = [] | |
| seen = 0 | |
| for dirpath, _, filenames in os.walk(root_dir): | |
| for fn in filenames: | |
| if not fn.lower().endswith(_IMG_EXTS): | |
| continue | |
| path = os.path.join(dirpath, fn) | |
| text = _caption_from_filename(fn) | |
| # 若提取失败就给空字符串;TextEncoder 会把空句子当作“无条件” | |
| items.append(CaptionItem(image_path=path, text=text)) | |
| seen += 1 | |
| if verbose and log_every > 0 and (seen % int(log_every) == 0): | |
| print(f"[scan] 已发现图片 {seen} 张...") | |
| items.sort(key=lambda x: x.image_path) | |
| if len(items) == 0: | |
| raise ValueError(f"目录下未找到图片:{root_dir}") | |
| if verbose: | |
| print(f"[scan] 完成:共 {len(items)} 张图片") | |
| return items | |