"""M1-④ 自动归类 读取 data/styles_meta.json，按规则分到 30 大类， '数字艺术' 进一步做二级细分；同时把 anime/manga 从插画拆出。 输出 data/catalog.json。 """ import argparse import json import re from collections import defaultdict from pathlib import Path ROOT = Path(__file__).resolve().parent.parent # ---------- 顶级大类规则（注意顺序：先匹配的优先） ---------- # 形如 (中文名, English, [关键字...])，关键字会以 lowercase substring 匹配 PRIMARY_RULES: list[tuple[str, str, list[str]]] = [ # 二次元单独抽出，必须在"插画"之前 ("二次元", "Anime/Manga", ["anime", "manga", "manhwa", "manhua", "chibi", "shojo", "shonen"]), # 数字艺术（之后会再做二级细分） ("数字艺术", "Digital Art", ["digital", " 3d", "3d ", "cgi", "vector", "render"]), # 插画 ("插画", "Illustration", ["illustration", "illustrative", "comic ", "comics", "cartoon", "storybook", "children's book"]), # 图像小说（必须在插画之后才不被吃掉？其实顺序：插画里不含 graphic novel，所以放哪都行） ("图像小说", "Graphic Novel", ["graphic novel"]), # 排版/字体 ("排版字体", "Typography", ["typography", "typographic", "lettering", "calligraphic poster"]), # 绘画 ("绘画", "Painting", ["painting", "oil ", " oil", "acrylic", "gouache", "tempera"]), # 水彩水墨 ("水彩水墨", "Watercolor/Ink", ["watercolor", "watercolour", "ink ", " ink", "sumi", "wash painting", "calligraph"]), # 版画与素描 ("版画素描", "Print/Sketch", ["lithograph", "etching", "engraving", "woodcut", "linocut", "sketch", "charcoal", "pencil"]), # 摄影 ("摄影", "Photography", ["photo", "photograph", "cinematic", "film noir", "noir"]), # 抽象 ("抽象", "Abstract", ["abstract", "abstraction", "geometric abstraction", "minimalism", "minimalist"]), # 印象/后印象 ("印象派", "Impressionism", ["impression", "post-impression", "pointillism"]), # 表现/超现实 ("表现超现实", "Surreal/Expressionism", ["expressionism", "expressive", "surreal", "dadaism", "psychedelic"]), # 文艺复兴/古典 ("古典", "Classical", ["renaissance", "baroque", "rococo", "neoclassic", "classical", "romanticism", "chiaroscuro"]), # 赛博朋克 ("赛博朋克", "Cyberpunk", ["cyberpunk", "futuristic", "retro-futur", "sci-fi", "synthwave", "vaporwave"]), # 蒸汽朋克 ("蒸汽朋克", "Steampunk", ["steampunk"]), # 波普/海报 ("波普海报", "Pop/Poster", ["pop ", "pop art", "poster", "propaganda", "graffiti", "street art"]), # 装饰艺术 ("装饰艺术", "Art Deco/Nouveau", ["deco", "nouveau", "arts and crafts"]), # 民俗与文化 ("民俗文化", "Folk/Cultural", ["folk", "tribal", "aboriginal", "ethnic", "ukiyo", "japanese ", "chinese ", "indian ", "african", "celtic", "mexican"]), # 幻想 ("幻想", "Fantasy", ["fantasy", "mythic", "mythological", "fairy"]), # 复古/做旧 ("复古", "Vintage/Retro", ["vintage", "retro", "antique", "victorian", "edwardian", "19th", "18th"]), # 风景/自然 ("风景自然", "Landscape/Nature", ["landscape", "botanical", "floral", "marine", "maritime", "seascape"]), # 肖像 ("肖像", "Portrait", ["portrait", "portraiture"]), # 建筑/室内 ("建筑室内", "Architecture", ["architectur", "interior", "urban", "cityscape"]), # 时尚 ("时尚", "Fashion", ["fashion", "editorial", "runway"]), # 拼贴/混合 ("拼贴混合", "Collage/Mixed", ["collage", "mixed media", "paper craft", "origami", "papercut"]), # 雕塑 ("雕塑", "Sculpture", ["sculpture", "bas-relief", "relief"]), # 暗黑/哥特 ("暗黑哥特", "Dark/Gothic", ["gothic", "dark ", "horror", "macabre", "dystop"]), # 写实/学院派（新增） ("写实学院", "Realism/Academic", ["realism", "realist", "academic", "academicism", "regionalism"]), # 现代/当代（新增） ("现代当代", "Modern/Contemporary", ["modernism", "modernist", "modern art", "contemporary"]), ] # ---------- 数字艺术二级细分 ---------- DIGITAL_SUB_RULES: list[tuple[str, str, list[str]]] = [ ("3D 动画/渲染", "3D Animation/Render", ["3d animation", "3d digital animation", "3d render", "3d rendering", "3d model", "3d ", " 3d"]), ("CGI/电影渲染", "CGI/Cinematic Render", ["cgi", "cinematic", "render", "rendering"]), ("矢量/扁平", "Vector/Flat", ["vector", "flat ", "flat design"]), ("数字绘画", "Digital Painting", ["digital painting"]), ("数字插画", "Digital Illustration", ["digital illustration", "digital illustrative"]), ("数字艺术其它", "Other Digital Art", []), # 兜底 ] def classify(prefix: str) -> tuple[str, str, str | None, str | None]: """返回 (cn, en, sub_cn, sub_en)。未命中返回 ('其它','Other',None,None)。""" p = (prefix or "").lower() for cn, en, kws in PRIMARY_RULES: if any(k in p for k in kws): if cn == "数字艺术": for scn, sen, skws in DIGITAL_SUB_RULES: if skws and any(k in p for k in skws): return cn, en, scn, sen return cn, en, "数字艺术其它", "Other Digital Art" return cn, en, None, None return "其它", "Other", None, None def main(): ap = argparse.ArgumentParser() ap.add_argument("--meta", default=str(ROOT / "data" / "styles_meta.json")) ap.add_argument("--out", default=str(ROOT / "data" / "catalog.json")) args = ap.parse_args() meta_path = Path(args.meta) out_path = Path(args.out) if not meta_path.exists(): raise SystemExit(f"找不到 {meta_path}, 请先运行 build_dataset.py") meta = json.loads(meta_path.read_text(encoding="utf-8")) # 主类 -> 子类 -> [style_key,...] catalog: dict[str, dict] = {} cat_count: dict[str, int] = defaultdict(int) sub_count: dict[str, int] = defaultdict(int) for key, info in meta.items(): prefix = info.get("style_prefix") or info.get("style_full") or key cn, en, scn, sen = classify(prefix) info["category"] = cn info["category_en"] = en info["sub_category"] = scn info["sub_category_en"] = sen cat_count[cn] += 1 bucket = catalog.setdefault(cn, { "name_cn": cn, "name_en": en, "count": 0, "subs": {}, "items": [], # 没有子类的直接平铺 }) bucket["count"] += 1 if scn: sub = bucket["subs"].setdefault(scn, { "name_cn": scn, "name_en": sen, "count": 0, "items": [], }) sub["count"] += 1 sub["items"].append(key) sub_count[f"{cn}/{scn}"] += 1 else: bucket["items"].append(key) # 写回 styles_meta.json (附加 category 字段) meta_path.write_text( json.dumps(meta, ensure_ascii=False, indent=2), encoding="utf-8" ) # 写出 catalog.json out_path.write_text( json.dumps(catalog, ensure_ascii=False, indent=2), encoding="utf-8" ) # 控制台输出统计 print(f"{'大类':<24}{'子类':<22}{'风格数':>8}") print("-" * 56) for cn, info in sorted(catalog.items(), key=lambda x: -x[1]["count"]): if info["subs"]: for scn, sub in sorted(info["subs"].items(), key=lambda x: -x[1]["count"]): print(f"{cn:<24}{scn:<22}{sub['count']:>8}") else: print(f"{cn:<24}{'':<22}{info['count']:>8}") print("-" * 56) total = sum(info["count"] for info in catalog.values()) print(f"{'合计':<46}{total:>8}") print(f"\n[完成] catalog.json -> {out_path}") print(f"[完成] meta 已附加 category 字段 -> {meta_path}") if __name__ == "__main__": main()