{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[COPY] /raid/workspace1/zxy/data/long_context/long_context_dataset_metadata.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/long_context_dataset_metadata.json\n",
      "[COPY] /raid/workspace1/zxy/Bagel_lora/data/all_data/x2edit_total.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/x2edit_total.json\n",
      "[COPY] /raid/workspace1/zxy/data/midjourney/midjourney_style_data_new.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/midjourney_style_data_new.json\n",
      "[COPY] /raid/workspace1/zxy/data/banana_scene/banana_scene_total.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/banana_scene_total.json\n",
      "[COPY] /raid/workspace1/zxy/data/banana_clothes_image/banana_change_clothes_dataset.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/banana_change_clothes_dataset.json\n",
      "[COPY] /raid/workspace1/zxy/data/banana_product_img/banana_model_product_dataset.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/banana_model_product_dataset.json\n",
      "[COPY] /raid/workspace1/zxy/data/Nano-150k/Nano_150k_json_complete/Nano_150k_merged_dedup.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Nano_150k_merged_dedup.json\n",
      "[COPY] /raid/workspace1/zxy/data/Echo-4o-Image/Echo_json_completed/hum_obj_complete_patched_v9_reindexed_dedup.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_obj_complete_patched_v9_reindexed_dedup.json\n",
      "[COPY] /raid/workspace1/zxy/data/Echo-4o-Image/Echo_json_completed/hum_obj_sce_complete_dedup.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_obj_sce_complete_dedup.json\n",
      "[COPY] /raid/workspace1/zxy/data/Echo-4o-Image/Echo_json_completed/hum_sce_complete_v2_reindexed_dedup.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_sce_complete_v2_reindexed_dedup.json\n",
      "[COPY] /raid/workspace1/zxy/data/Echo-4o-Image/Echo_json_completed/human_complete_patched_v2_reindexed_dedup.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/human_complete_patched_v2_reindexed_dedup.json\n",
      "[COPY] /raid/workspace1/zxy/data/Echo-4o-Image/Echo_json_completed/obj_sce_complete_reindexed_dedup.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/obj_sce_complete_reindexed_dedup.json\n",
      "[COPY] /raid/workspace1/zxy/data/Echo-4o-Image/Echo_json_completed/object_complete2_reindexed_dedup.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/object_complete2_reindexed_dedup.json\n",
      "[COPY] /raid/workspace1/zxy/data/Echo-4o-Image/Echo_json_completed/other_complete_dedup.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/other_complete_dedup.json\n",
      "[COPY] /raid/workspace1/zxy/data/nano_personalized/nano_customized.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/nano_customized.json\n",
      "[COPY] /raid/workspace1/zxy/data/nano_orc/nano_orc_final.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/nano_orc_final.json\n",
      "[COPY] /raid/workspace1/zxy/data/more_multi_condition/more_multi.json -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/more_multi.json\n",
      "[DONE] All requested JSONs copied under: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken\n"
     ]
    }
   ],
   "source": [
    "#!/usr/bin/env python3\n",
    "# -*- coding: utf-8 -*-\n",
    "\n",
    "import os\n",
    "import shutil\n",
    "from pathlib import Path\n",
    "from typing import List, Tuple\n",
    "\n",
    "# ======================\n",
    "# Inputs: (json_path_or_dir, data_root, is_dir)  ← 按你给的列表原样放入\n",
    "# ======================\n",
    "INPUT_SPECS: List[Tuple[str, str, bool]] = [\n",
    "    (\"/fi-lib/workspace/zxy/data/long_context/long_context_dataset_metadata.json\",\n",
    "     \"data:/fi-lib/share_data/long_context\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace1/zxy/Bagel_lora/data/all_data/x2edit_total_new.json\",\n",
    "     \"data:/fi-lib/share_data/X2Edit_data\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/midjourney/midjourney_style_data_new.json\",\n",
    "     \"data:/fi-lib/share_data\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/banana_scene/banana_scene_total.json\",\n",
    "     \"data:/fi-lib/share_data/banana_scene\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/banana_clothes_image/banana_change_clothes_dataset.json\",\n",
    "     \"data:/fi-lib/workspace/zxy/data\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/banana_product_img/banana_model_product_dataset.json\",\n",
    "     \"data:/fi-lib/workspace/zxy/data\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/Nano-150k/Nano_150k_json_complete/Nano_150k_merged_dedup.json\",\n",
    "     \"data:/fi-lib/share_data/Nano-consistent-150k\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/Echo-4o-Image/Echo_json_completed\",\n",
    "     \"data:/fi-lib/share_data\".replace(\"data:\", \"\"),\n",
    "     True),  # directory of JSONs\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/nano_personalized/nano_customized.json\",\n",
    "     \"data:/fi-lib/workspace/zxy/data/nano_personalized\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/nano_orc/nano_orc_final.json\",\n",
    "     \"data:/fi-lib/workspace/zxy/data/nano_orc\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "\n",
    "    (\"/fi-lib/workspace/zxy/data/more_multi_condition/more_multi.json\",\n",
    "     \"data:/fi-lib/workspace/zxy/data/more_multi_condition/valid\".replace(\"data:\", \"\"),\n",
    "     False),\n",
    "]\n",
    "\n",
    "DEST_BASE = Path(\"/fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken\")\n",
    "DEST_BASE.mkdir(parents=True, exist_ok=True)\n",
    "\n",
    "def copy_file(src: Path, dst_dir: Path):\n",
    "    dst_dir.mkdir(parents=True, exist_ok=True)\n",
    "    dst_path = dst_dir / src.name\n",
    "    shutil.copy2(src, dst_path)\n",
    "    print(f\"[COPY] {src} -> {dst_path}\")\n",
    "\n",
    "def copy_dir_jsons(src_dir: Path, dst_base: Path):\n",
    "    \"\"\"\n",
    "    递归复制 src_dir 下的所有 .json，保持 src_dir 的目录名与层级：\n",
    "    例如 src_dir=.../Echo_json_completed\n",
    "      -> 复制到 dst_base/Echo_json_completed/...(子目录结构)/xxx.json\n",
    "    \"\"\"\n",
    "    if not src_dir.exists():\n",
    "        print(f\"[WARN] dir not found: {src_dir}\")\n",
    "        return\n",
    "    rel_root_name = src_dir.name\n",
    "    for p in src_dir.rglob(\"*.json\"):\n",
    "        # 在目标中保持相对 src_dir 的层级\n",
    "        rel = p.relative_to(src_dir)\n",
    "        dst_dir = dst_base / rel_root_name / rel.parent\n",
    "        dst_dir.mkdir(parents=True, exist_ok=True)\n",
    "        dst_path = dst_dir / p.name\n",
    "        shutil.copy2(p, dst_path)\n",
    "        print(f\"[COPY] {p} -> {dst_path}\")\n",
    "\n",
    "def main():\n",
    "    for path_str, _data_root, is_dir in INPUT_SPECS:\n",
    "        src_path = Path(path_str)\n",
    "        if is_dir:\n",
    "            copy_dir_jsons(src_path, DEST_BASE)\n",
    "        else:\n",
    "            if not src_path.exists():\n",
    "                print(f\"[WARN] file not found: {src_path}\")\n",
    "                continue\n",
    "            copy_file(src_path, DEST_BASE)\n",
    "\n",
    "    print(f\"[DONE] All requested JSONs copied under: {DEST_BASE}\")\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/Nano_150k_merged_dedup.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/Nano_150k_merged_dedup.json  (删掉 6 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/banana_model_product_dataset.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/banana_model_product_dataset.json  (删掉 2 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/long_context_dataset_metadata.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/long_context_dataset_metadata.json  (删掉 7 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/nano_customized.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/nano_customized.json  (删掉 1 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/banana_change_clothes_dataset.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/banana_change_clothes_dataset.json  (删掉 2 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/midjourney_style_data_new.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/midjourney_style_data_new.json  (删掉 0 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/more_multi.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/more_multi.json  (删掉 5 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/x2edit_total.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/x2edit_total.json  (删掉 12 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/banana_scene_total.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/banana_scene_total.json  (删掉 2 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/nano_orc_final.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/nano_orc_final.json  (删掉 1 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/x2edit_total_new.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/x2edit_total_new.json  (删掉 12 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/other_complete_dedup.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/other_complete_dedup.json  (删掉 4 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/object_complete2_reindexed_dedup.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/object_complete2_reindexed_dedup.json  (删掉 3 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/human_complete_patched_v2_reindexed_dedup.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/human_complete_patched_v2_reindexed_dedup.json  (删掉 2 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_obj_complete_patched_v9_reindexed_dedup.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/hum_obj_complete_patched_v9_reindexed_dedup.json  (删掉 4 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/obj_sce_complete_reindexed_dedup.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/obj_sce_complete_reindexed_dedup.json  (删掉 4 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_obj_sce_complete_dedup.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/hum_obj_sce_complete_dedup.json  (删掉 4 类token)\n",
      "[OK] 处理并保存: /fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_sce_complete_v2_reindexed_dedup.json  ->  /fi-lib/workspace/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/hum_sce_complete_v2_reindexed_dedup.json  (删掉 4 类token)\n",
      "\n",
      "===== 汇总 =====\n",
      "共处理 JSON 文件：18\n",
      "删除过的 tokens（不重复）：\n",
      "  - <background>\n",
      "  - <clothing>\n",
      "  - <emotion>\n",
      "  - <environment>\n",
      "  - <icon>\n",
      "  - <id>\n",
      "  - <lighting>\n",
      "  - <perspective>\n",
      "  - <pose>\n",
      "  - <subject>\n",
      "  - <text>\n",
      "  - <texture>\n"
     ]
    }
   ],
   "source": [
    "#!/usr/bin/env python3\n",
    "# -*- coding: utf-8 -*-\n",
    "\n",
    "import os\n",
    "import json\n",
    "import re\n",
    "from pathlib import Path\n",
    "from typing import Any, Dict, List, Set\n",
    "\n",
    "SRC_BASE = Path(\"/fi-lib/workspace/zxy/data/all_combine/all_combine_specialtoken\")\n",
    "DST_BASE = Path(\"/fi-lib/workspace/zxy/data/all_combine/all_combine_notoken\")\n",
    "\n",
    "# 要移除的 special tokens（精确匹配，区分大小写）\n",
    "TOKENS = [\n",
    "    \"<subject>\", \"<id>\", \"<icon>\", \"<text>\", \"<pose>\", \"<layout>\",\n",
    "    \"<perspective>\", \"<lighting>\", \"<background>\", \"<environment>\",\n",
    "    \"<emotion>\", \"<clothing>\", \"<texture>\", \"<REF_AIM>\", \"<EDIT_BASE>\"\n",
    "]\n",
    "\n",
    "TARGET_PROMPT_KEYS = {\"prompt\", \"prompt_used\"}  # 处理这些键的字符串\n",
    "\n",
    "# 预编译：为每个 token 准备正则（直接字面量替换）\n",
    "TOKEN_PATTERNS = [(t, re.compile(re.escape(t))) for t in TOKENS]\n",
    "\n",
    "def clean_text(s: str, deleted_set: Set[str]) -> str:\n",
    "    \"\"\"从字符串中移除所有 special tokens，并做轻度空白规范化。\"\"\"\n",
    "    if not isinstance(s, str):\n",
    "        return s\n",
    "    out = s\n",
    "    for token, pat in TOKEN_PATTERNS:\n",
    "        if token in out:\n",
    "            deleted_set.add(token)\n",
    "            out = pat.sub(\"\", out)\n",
    "    # 轻度清理多余空格\n",
    "    out = re.sub(r\"\\s{2,}\", \" \", out).strip()\n",
    "    return out\n",
    "\n",
    "def traverse_and_clean(obj: Any, deleted_set: Set[str]) -> Any:\n",
    "    \"\"\"\n",
    "    递归遍历：仅对 'prompt' / 'prompt_used' 的字符串做替换；\n",
    "    其他键不改，子结构继续深入。\n",
    "    \"\"\"\n",
    "    if isinstance(obj, dict):\n",
    "        new_d = {}\n",
    "        for k, v in obj.items():\n",
    "            if k in TARGET_PROMPT_KEYS and isinstance(v, str):\n",
    "                new_d[k] = clean_text(v, deleted_set)\n",
    "            else:\n",
    "                new_d[k] = traverse_and_clean(v, deleted_set)\n",
    "        return new_d\n",
    "    elif isinstance(obj, list):\n",
    "        return [traverse_and_clean(x, deleted_set) for x in obj]\n",
    "    else:\n",
    "        return obj  # 其他类型原样返回\n",
    "\n",
    "def process_one_json(src_path: Path, dst_path: Path, deleted_set_global: Set[str]) -> None:\n",
    "    try:\n",
    "        with src_path.open(\"r\", encoding=\"utf-8\") as f:\n",
    "            data = json.load(f)\n",
    "    except Exception as e:\n",
    "        print(f\"[SKIP] 读取失败: {src_path} ({e})\")\n",
    "        return\n",
    "\n",
    "    deleted_local: Set[str] = set()\n",
    "    new_data = traverse_and_clean(data, deleted_local)\n",
    "\n",
    "    # 合并到全局集合\n",
    "    deleted_set_global.update(deleted_local)\n",
    "\n",
    "    dst_path.parent.mkdir(parents=True, exist_ok=True)\n",
    "    with dst_path.open(\"w\", encoding=\"utf-8\") as f:\n",
    "        json.dump(new_data, f, ensure_ascii=False, indent=2)\n",
    "\n",
    "    print(f\"[OK] 处理并保存: {src_path}  ->  {dst_path}  (删掉 {len(deleted_local)} 类token)\")\n",
    "\n",
    "def main():\n",
    "    if not SRC_BASE.exists():\n",
    "        raise FileNotFoundError(f\"源目录不存在：{SRC_BASE}\")\n",
    "    DST_BASE.mkdir(parents=True, exist_ok=True)\n",
    "\n",
    "    deleted_tokens_global: Set[str] = set()\n",
    "    count_files = 0\n",
    "\n",
    "    for p in SRC_BASE.rglob(\"*.json\"):\n",
    "        # 在目标中保持相对路径与文件名\n",
    "        rel = p.relative_to(SRC_BASE)\n",
    "        dst_p = DST_BASE / rel\n",
    "        process_one_json(p, dst_p, deleted_tokens_global)\n",
    "        count_files += 1\n",
    "\n",
    "    print(\"\\n===== 汇总 =====\")\n",
    "    print(f\"共处理 JSON 文件：{count_files}\")\n",
    "    if deleted_tokens_global:\n",
    "        print(\"删除过的 tokens（不重复）：\")\n",
    "        for t in sorted(deleted_tokens_global):\n",
    "            print(\"  -\", t)\n",
    "    else:\n",
    "        print(\"未发现需要删除的 tokens。\")\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/long_context_dataset_metadata.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/long_context_dataset_metadata.json  (删掉 7 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/x2edit_total.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/x2edit_total.json  (删掉 12 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/midjourney_style_data_new.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/midjourney_style_data_new.json  (删掉 0 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/banana_scene_total.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/banana_scene_total.json  (删掉 2 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/banana_change_clothes_dataset.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/banana_change_clothes_dataset.json  (删掉 2 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/banana_model_product_dataset.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/banana_model_product_dataset.json  (删掉 2 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Nano_150k_merged_dedup.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/Nano_150k_merged_dedup.json  (删掉 6 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/nano_customized.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/nano_customized.json  (删掉 1 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/nano_orc_final.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/nano_orc_final.json  (删掉 1 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/more_multi.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/more_multi.json  (删掉 5 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_obj_complete_patched_v9_reindexed_dedup.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/hum_obj_complete_patched_v9_reindexed_dedup.json  (删掉 4 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_obj_sce_complete_dedup.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/hum_obj_sce_complete_dedup.json  (删掉 4 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/hum_sce_complete_v2_reindexed_dedup.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/hum_sce_complete_v2_reindexed_dedup.json  (删掉 4 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/human_complete_patched_v2_reindexed_dedup.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/human_complete_patched_v2_reindexed_dedup.json  (删掉 3 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/obj_sce_complete_reindexed_dedup.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/obj_sce_complete_reindexed_dedup.json  (删掉 4 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/object_complete2_reindexed_dedup.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/object_complete2_reindexed_dedup.json  (删掉 3 类token)\n",
      "[OK] 处理并保存: /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Echo_json_completed/other_complete_dedup.json  ->  /raid/workspace1/zxy/data/all_combine/all_combine_notoken/Echo_json_completed/other_complete_dedup.json  (删掉 4 类token)\n",
      "\n",
      "===== 汇总 =====\n",
      "共处理 JSON 文件：17\n",
      "\n",
      "出现过并被删除的 tokens（含总删除次数）：\n",
      "  - <background>: 37589\n",
      "  - <clothing>: 19196\n",
      "  - <emotion>: 759\n",
      "  - <environment>: 16071\n",
      "  - <icon>: 1\n",
      "  - <id>: 154379\n",
      "  - <lighting>: 4252\n",
      "  - <perspective>: 1416\n",
      "  - <pose>: 28767\n",
      "  - <subject>: 138356\n",
      "  - <text>: 467\n",
      "  - <texture>: 105\n",
      "\n",
      "未出现（未删除）的 tokens：\n",
      "  - <EDIT_BASE>\n",
      "  - <REF_AIM>\n",
      "  - <layout>\n"
     ]
    }
   ],
   "source": [
    "#!/usr/bin/env python3\n",
    "# -*- coding: utf-8 -*-\n",
    "\n",
    "import os\n",
    "import json\n",
    "import re\n",
    "from pathlib import Path\n",
    "from typing import Any, Dict, List, Set\n",
    "\n",
    "SRC_BASE = Path(\"/raid/workspace1/zxy/data/all_combine/all_combine_specialtoken\")\n",
    "DST_BASE = Path(\"/raid/workspace1/zxy/data/all_combine/all_combine_notoken\")\n",
    "\n",
    "TOKENS = [\n",
    "    \"<subject>\", \"<id>\", \"<icon>\", \"<text>\", \"<pose>\", \"<layout>\",\n",
    "    \"<perspective>\", \"<lighting>\", \"<background>\", \"<environment>\",\n",
    "    \"<emotion>\", \"<clothing>\", \"<texture>\", \"<REF_AIM>\", \"<EDIT_BASE>\"\n",
    "]\n",
    "TARGET_PROMPT_KEYS = {\"prompt\", \"prompt_used\"}\n",
    "\n",
    "TOKEN_PATTERNS = [(t, re.compile(re.escape(t))) for t in TOKENS]\n",
    "\n",
    "# 统计每个 token 被删除的次数\n",
    "token_hit_counts: Dict[str, int] = {t: 0 for t in TOKENS}\n",
    "\n",
    "def clean_text(s: str, deleted_set: Set[str]) -> str:\n",
    "    if not isinstance(s, str):\n",
    "        return s\n",
    "    out = s\n",
    "    for token, pat in TOKEN_PATTERNS:\n",
    "        # 统计本次字符串中出现的次数\n",
    "        hits = len(pat.findall(out))\n",
    "        if hits > 0:\n",
    "            deleted_set.add(token)\n",
    "            token_hit_counts[token] += hits\n",
    "            out = pat.sub(\"\", out)\n",
    "    out = re.sub(r\"\\s{2,}\", \" \", out).strip()\n",
    "    return out\n",
    "\n",
    "def traverse_and_clean(obj: Any, deleted_set: Set[str]) -> Any:\n",
    "    if isinstance(obj, dict):\n",
    "        new_d = {}\n",
    "        for k, v in obj.items():\n",
    "            if k in TARGET_PROMPT_KEYS and isinstance(v, str):\n",
    "                new_d[k] = clean_text(v, deleted_set)\n",
    "            else:\n",
    "                new_d[k] = traverse_and_clean(v, deleted_set)\n",
    "        return new_d\n",
    "    elif isinstance(obj, list):\n",
    "        return [traverse_and_clean(x, deleted_set) for x in obj]\n",
    "    else:\n",
    "        return obj\n",
    "\n",
    "def process_one_json(src_path: Path, dst_path: Path, deleted_set_global: Set[str]) -> None:\n",
    "    try:\n",
    "        with src_path.open(\"r\", encoding=\"utf-8\") as f:\n",
    "            data = json.load(f)\n",
    "    except Exception as e:\n",
    "        print(f\"[SKIP] 读取失败: {src_path} ({e})\")\n",
    "        return\n",
    "\n",
    "    deleted_local: Set[str] = set()\n",
    "    new_data = traverse_and_clean(data, deleted_local)\n",
    "    deleted_set_global.update(deleted_local)\n",
    "\n",
    "    dst_path.parent.mkdir(parents=True, exist_ok=True)\n",
    "    with dst_path.open(\"w\", encoding=\"utf-8\") as f:\n",
    "        json.dump(new_data, f, ensure_ascii=False, indent=2)\n",
    "\n",
    "    print(f\"[OK] 处理并保存: {src_path}  ->  {dst_path}  (删掉 {len(deleted_local)} 类token)\")\n",
    "\n",
    "def main():\n",
    "    if not SRC_BASE.exists():\n",
    "        raise FileNotFoundError(f\"源目录不存在：{SRC_BASE}\")\n",
    "    DST_BASE.mkdir(parents=True, exist_ok=True)\n",
    "\n",
    "    deleted_tokens_global: Set[str] = set()\n",
    "    count_files = 0\n",
    "\n",
    "    for p in SRC_BASE.rglob(\"*.json\"):\n",
    "        rel = p.relative_to(SRC_BASE)\n",
    "        dst_p = DST_BASE / rel\n",
    "        process_one_json(p, dst_p, deleted_tokens_global)\n",
    "        count_files += 1\n",
    "\n",
    "    print(\"\\n===== 汇总 =====\")\n",
    "    print(f\"共处理 JSON 文件：{count_files}\")\n",
    "\n",
    "    used_tokens = sorted(deleted_tokens_global)\n",
    "    unused_tokens = sorted(set(TOKENS) - deleted_tokens_global)\n",
    "\n",
    "    print(\"\\n出现过并被删除的 tokens（含总删除次数）：\")\n",
    "    if used_tokens:\n",
    "        for t in used_tokens:\n",
    "            print(f\"  - {t}: {token_hit_counts.get(t, 0)}\")\n",
    "    else:\n",
    "        print(\"  （无）\")\n",
    "\n",
    "    print(\"\\n未出现（未删除）的 tokens：\")\n",
    "    if unused_tokens:\n",
    "        for t in unused_tokens:\n",
    "            print(f\"  - {t}\")\n",
    "    else:\n",
    "        print(\"  （全部都出现过）\")\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[OK] Processed 123264 items; modified 15508 items containing '<style>' under method 'nano150k_sweet_headshot'.\n",
      "[SAVE] -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Nano_150k_merged_dedup.json\n"
     ]
    }
   ],
   "source": [
    "#!/usr/bin/env python3\n",
    "# -*- coding: utf-8 -*-\n",
    "\n",
    "import os, json, re\n",
    "from pathlib import Path\n",
    "\n",
    "IN_PATH  = \"/raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/Nano_150k_merged_dedup.json\"\n",
    "OUT_PATH = IN_PATH  # 如需另存，改成别的路径\n",
    "\n",
    "METHOD_TARGET = \"nano150k_sweet_headshot\"\n",
    "STYLE_TOKEN   = \"<style>\"\n",
    "\n",
    "def clean_style_token(text: str) -> str:\n",
    "    if not isinstance(text, str):\n",
    "        return text\n",
    "    # 删除 <style>，然后把多余空格合并\n",
    "    out = text.replace(STYLE_TOKEN, \"\")\n",
    "    out = re.sub(r\"\\s{2,}\", \" \", out).strip()\n",
    "    return out\n",
    "\n",
    "def process_sample(sample: dict) -> bool:\n",
    "    \"\"\"\n",
    "    返回是否有修改\n",
    "    \"\"\"\n",
    "    changed = False\n",
    "    if sample.get(\"data_prep_method\") != METHOD_TARGET:\n",
    "        return False\n",
    "\n",
    "    if \"prompt_used\" in sample and isinstance(sample[\"prompt_used\"], str):\n",
    "        newp = clean_style_token(sample[\"prompt_used\"])\n",
    "        if newp != sample[\"prompt_used\"]:\n",
    "            sample[\"prompt_used\"] = newp\n",
    "            changed = True\n",
    "\n",
    "    # 若也存在 prompt 字段，一并处理\n",
    "    if \"prompt\" in sample and isinstance(sample[\"prompt\"], str):\n",
    "        newp2 = clean_style_token(sample[\"prompt\"])\n",
    "        if newp2 != sample[\"prompt\"]:\n",
    "            sample[\"prompt\"] = newp2\n",
    "            changed = True\n",
    "\n",
    "    return changed\n",
    "\n",
    "def main():\n",
    "    if not os.path.exists(IN_PATH):\n",
    "        raise FileNotFoundError(IN_PATH)\n",
    "\n",
    "    with open(IN_PATH, \"r\", encoding=\"utf-8\") as f:\n",
    "        data = json.load(f)\n",
    "\n",
    "    # 支持 {\"sample_x\": {...}} 或 list 结构\n",
    "    pairs = []\n",
    "    if isinstance(data, dict):\n",
    "        pairs = [(k, data[k]) for k in data.keys()]\n",
    "    elif isinstance(data, list):\n",
    "        pairs = list(enumerate(data))\n",
    "    else:\n",
    "        raise TypeError(f\"Unsupported top-level type: {type(data)}\")\n",
    "\n",
    "    total = len(pairs)\n",
    "    touched = 0\n",
    "\n",
    "    for _, sample in pairs:\n",
    "        if isinstance(sample, dict):\n",
    "            if process_sample(sample):\n",
    "                touched += 1\n",
    "\n",
    "    # 保存\n",
    "    Path(OUT_PATH).parent.mkdir(parents=True, exist_ok=True)\n",
    "    with open(OUT_PATH, \"w\", encoding=\"utf-8\") as f:\n",
    "        json.dump(data, f, ensure_ascii=False, indent=2)\n",
    "\n",
    "    print(f\"[OK] Processed {total} items; modified {touched} items containing '{STYLE_TOKEN}' \"\n",
    "          f\"under method '{METHOD_TARGET}'.\")\n",
    "    print(f\"[SAVE] -> {OUT_PATH}\")\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[OK] 完成：已在 3543 处删除第 2 次出现的 '<subject>'\n",
      "[SAVE] -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/nano_customized.json\n"
     ]
    }
   ],
   "source": [
    "#!/usr/bin/env python3\n",
    "# -*- coding: utf-8 -*-\n",
    "\n",
    "import os, json, re\n",
    "from pathlib import Path\n",
    "from typing import Any, Dict\n",
    "\n",
    "IN_PATH  = \"/raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/nano_customized.json\"\n",
    "OUT_PATH = IN_PATH  # 如需另存为新文件，改成其他路径\n",
    "\n",
    "TARGET_PROMPT_KEYS = {\"prompt\", \"prompt_used\"}\n",
    "TOKEN = \"<subject>\"\n",
    "PAT = re.compile(re.escape(TOKEN))\n",
    "\n",
    "def remove_nth_occurrence(s: str, token_pat: re.Pattern, n: int) -> str:\n",
    "    \"\"\"\n",
    "    删除字符串中第 n 次出现的 token（基于正则精确匹配）。\n",
    "    若出现次数 < n，则不改动。\n",
    "    \"\"\"\n",
    "    if not isinstance(s, str):\n",
    "        return s\n",
    "    matches = list(token_pat.finditer(s))\n",
    "    if len(matches) < n:\n",
    "        return s\n",
    "    m = matches[n-1]\n",
    "    # 删除该次匹配\n",
    "    return s[:m.start()] + s[m.end():]\n",
    "\n",
    "def process_obj(obj: Any, counter: Dict[str, int]) -> Any:\n",
    "    \"\"\"\n",
    "    递归遍历：仅对 'prompt' / 'prompt_used' 的字符串删第2个 <subject>。\n",
    "    其他键保持不变。\n",
    "    \"\"\"\n",
    "    if isinstance(obj, dict):\n",
    "        out = {}\n",
    "        for k, v in obj.items():\n",
    "            if k in TARGET_PROMPT_KEYS and isinstance(v, str):\n",
    "                new_v = remove_nth_occurrence(v, PAT, 2)\n",
    "                if new_v != v:\n",
    "                    counter[\"modified\"] += 1\n",
    "                out[k] = new_v\n",
    "            else:\n",
    "                out[k] = process_obj(v, counter)\n",
    "        return out\n",
    "    elif isinstance(obj, list):\n",
    "        return [process_obj(x, counter) for x in obj]\n",
    "    else:\n",
    "        return obj\n",
    "\n",
    "def main():\n",
    "    if not os.path.exists(IN_PATH):\n",
    "        raise FileNotFoundError(IN_PATH)\n",
    "\n",
    "    with open(IN_PATH, \"r\", encoding=\"utf-8\") as f:\n",
    "        data = json.load(f)\n",
    "\n",
    "    counter = {\"modified\": 0}\n",
    "    new_data = process_obj(data, counter)\n",
    "\n",
    "    Path(OUT_PATH).parent.mkdir(parents=True, exist_ok=True)\n",
    "    with open(OUT_PATH, \"w\", encoding=\"utf-8\") as f:\n",
    "        json.dump(new_data, f, ensure_ascii=False, indent=2)\n",
    "\n",
    "    print(f\"[OK] 完成：已在 {counter['modified']} 处删除第 2 次出现的 '{TOKEN}'\")\n",
    "    print(f\"[SAVE] -> {OUT_PATH}\")\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[OK] Rewrote prompts. Modified 469 samples.\n",
      "[SAVE] -> /raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/x2edit_total_new.json\n"
     ]
    }
   ],
   "source": [
    "#!/usr/bin/env python3\n",
    "# -*- coding: utf-8 -*-\n",
    "\n",
    "import os, json, re\n",
    "from pathlib import Path\n",
    "\n",
    "IN_PATH  = \"/raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/x2edit_total.json\"\n",
    "OUT_PATH = \"/raid/workspace1/zxy/data/all_combine/all_combine_specialtoken/x2edit_total_new.json\"  # 如需另存为新文件，改成其它路径\n",
    "\n",
    "# 预编译正则\n",
    "RE_TRAIL_IN = re.compile(r\"\\s+in\\s+<image_1>\\s*([.!?])?$\")  # 句末 in <image_1>（可带收尾标点）\n",
    "RE_ZOOM = re.compile(r\"\\bZoom\\s+(in|out)\\s*<image_1>\", flags=0)  # Zoom in/out 紧随 <image_1>\n",
    "RE_LOGO_FROM = re.compile(r\"logo from\")  # 区分大小写，按你的示例\n",
    "RE_IN_ANY = re.compile(r\"\\s+in\\s+<image_1>\")  # 句中任意 in <image_1>\n",
    "\n",
    "def strip_trailing_in_image1(s: str) -> str:\n",
    "    \"\"\"去掉句末的 ' in <image_1>'，保留结尾标点。\"\"\"\n",
    "    m = RE_TRAIL_IN.search(s)\n",
    "    if not m:\n",
    "        return s\n",
    "    punct = m.group(1) or \"\"\n",
    "    return RE_TRAIL_IN.sub(punct, s)\n",
    "\n",
    "def ensure_in_image1_at_end(s: str) -> str:\n",
    "    \"\"\"在句末追加 ' in <image_1>'；若已有则不重复。\"\"\"\n",
    "    # 已经以 in <image_1>（可带标点）结尾则不加\n",
    "    if re.search(r\"\\s+in\\s+<image_1>\\s*([.!?])?$\", s):\n",
    "        return s\n",
    "    # 带标点时插在标点前\n",
    "    m = re.search(r\"([.!?])\\s*$\", s)\n",
    "    if m:\n",
    "        pos = m.start(1)\n",
    "        return s[:pos] + \" in <image_1>\" + s[pos:]\n",
    "    else:\n",
    "        return s.rstrip() + \" in <image_1>\"\n",
    "\n",
    "def rewrite_prompt(p: str) -> str:\n",
    "    if not isinstance(p, str):\n",
    "        return p\n",
    "    out = p\n",
    "\n",
    "    # 1) 专门规则：logo from —— 先去掉句中所有 in <image_1>，再改为 \"logo in <image_1> from\"\n",
    "    if \"logo from\" in out:\n",
    "        out = RE_IN_ANY.sub(\"\", out)\n",
    "        out = RE_LOGO_FROM.sub(\"logo in <image_1> from\", out)\n",
    "\n",
    "    # 2) \"' to '\" 结构：去掉句末 in <image_1>，再替换为 \"' in <image_1> to '\"\n",
    "    if \"' to '\" in out:\n",
    "        out = strip_trailing_in_image1(out)\n",
    "        out = out.replace(\"' to '\", \"' in <image_1> to '\")\n",
    "\n",
    "    # 3) \"' from \" 结构：去掉句末 in <image_1>，再替换为 \"' in <image_1> from \"\n",
    "    if \"' from \" in out:\n",
    "        out = strip_trailing_in_image1(out)\n",
    "        out = out.replace(\"' from \", \"' in <image_1> from \")\n",
    "\n",
    "    # 4) \"' with \" 结构：去掉句末 in <image_1>，再替换为 \"' in <image_1> with \"\n",
    "    if \"' with \" in out:\n",
    "        out = strip_trailing_in_image1(out)\n",
    "        out = out.replace(\"' with \", \"' in <image_1> with \")\n",
    "\n",
    "    # 5) Zoom in/out：删除紧跟的 <image_1>，并把 \" in <image_1>\" 放到句尾（标点前）\n",
    "    if \"Zoom in <image_1>\" in out or \"Zoom out <image_1>\" in out:\n",
    "        out = RE_ZOOM.sub(lambda m: f\"Zoom {m.group(1)}\", out)\n",
    "        out = ensure_in_image1_at_end(out)\n",
    "\n",
    "    return out\n",
    "\n",
    "def main():\n",
    "    if not os.path.exists(IN_PATH):\n",
    "        raise FileNotFoundError(IN_PATH)\n",
    "\n",
    "    with open(IN_PATH, \"r\", encoding=\"utf-8\") as f:\n",
    "        data = json.load(f)\n",
    "\n",
    "    # 支持 {\"sample_xxx\": {...}} 或 list 结构\n",
    "    if isinstance(data, dict):\n",
    "        items = data.items()\n",
    "        def assign(k, v): data[k] = v\n",
    "    elif isinstance(data, list):\n",
    "        items = enumerate(data)\n",
    "        def assign(k, v): data[k] = v\n",
    "    else:\n",
    "        raise TypeError(f\"Unsupported top-level type: {type(data)}\")\n",
    "\n",
    "    touched = 0\n",
    "    for k, sample in items:\n",
    "        if not isinstance(sample, dict):\n",
    "            continue\n",
    "        p = sample.get(\"prompt_used\")\n",
    "        if isinstance(p, str):\n",
    "            new_p = rewrite_prompt(p)\n",
    "            if new_p != p:\n",
    "                sample[\"prompt_used\"] = new_p\n",
    "                touched += 1\n",
    "                assign(k, sample)\n",
    "\n",
    "    # 保存（原地覆盖）\n",
    "    Path(OUT_PATH).parent.mkdir(parents=True, exist_ok=True)\n",
    "    with open(OUT_PATH, \"w\", encoding=\"utf-8\") as f:\n",
    "        json.dump(data, f, ensure_ascii=False, indent=2)\n",
    "\n",
    "    print(f\"[OK] Rewrote prompts. Modified {touched} samples.\")\n",
    "    print(f\"[SAVE] -> {OUT_PATH}\")\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "=== Processing sample #1 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano_customized/test_sample112_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/1/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano_customized/test_sample112_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/1/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano_customized/test_sample112_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/1/instruction_token.txt\n",
      "=== Processing sample #2 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample1_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/2/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample1_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/2/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample1_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/2/instruction_token.txt\n",
      "=== Processing sample #3 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample6_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/3/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample6_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/3/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample6_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/3/instruction_token.txt\n",
      "=== Processing sample #4 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample22_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/4/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample22_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/4/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample22_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/4/instruction_token.txt\n",
      "=== Processing sample #5 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_background/test_sample0_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/5/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_background/test_sample0_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/5/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_background/test_sample0_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/5/instruction_token.txt\n",
      "=== Processing sample #6 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample3_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/6/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample3_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/6/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample3_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/6/instruction_token.txt\n",
      "=== Processing sample #7 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample5_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/7/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample5_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/7/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample5_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/7/instruction_token.txt\n",
      "=== Processing sample #8 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_Times_Change/test_sample18_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/8/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_Times_Change/test_sample18_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/8/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_Times_Change/test_sample18_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/8/instruction_token.txt\n",
      "=== Processing sample #9 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample11_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/9/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample11_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/9/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample11_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/9/instruction_token.txt\n",
      "=== Processing sample #10 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample71_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/10/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample71_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/10/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample71_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/10/instruction_token.txt\n",
      "=== Processing sample #11 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample73_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/11/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample73_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/11/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample73_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/11/instruction_token.txt\n",
      "=== Processing sample #12 ===\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample80_img_0.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/12/orig_000.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample80_img_gt.jpg -> /fi-lib/workspace/zxy/data/A_benchmark/edit/12/gt.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample80_instruction.txt -> /fi-lib/workspace/zxy/data/A_benchmark/edit/12/instruction_token.txt\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import shutil\n",
    "from pathlib import Path\n",
    "\n",
    "# ===== 手动列出的 generated 图片路径 =====\n",
    "GENERATED_PATHS = [\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano_customized/test_sample112_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample1_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample6_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample22_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_background/test_sample0_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample3_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample5_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_Times_Change/test_sample18_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample11_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample71_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample73_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample80_img_generated.jpg\",\n",
    "]\n",
    "\n",
    "# ===== 目标根目录 =====\n",
    "OUT_ROOT = Path(\"/fi-lib/workspace/zxy/data/A_benchmark/edit\")\n",
    "\n",
    "\n",
    "def copy_one_sample(src_generated_path: str, dst_idx: int):\n",
    "    src_generated = Path(src_generated_path)\n",
    "    if not src_generated.is_file():\n",
    "        print(f\"[WARN] generated not found: {src_generated}\")\n",
    "        return\n",
    "\n",
    "    # 前缀去掉 \"_img_generated.jpg\"\n",
    "    name = src_generated.name  # e.g. test_sample80_img_generated.jpg\n",
    "    if not name.endswith(\"_img_generated.jpg\"):\n",
    "        print(f\"[WARN] Unexpected filename pattern: {src_generated}\")\n",
    "        return\n",
    "\n",
    "    prefix = name.replace(\"_img_generated.jpg\", \"\")  # e.g. test_sample80\n",
    "    src_dir = src_generated.parent\n",
    "\n",
    "    # 构造其他相关文件路径\n",
    "    src_0 = src_dir / f\"{prefix}_img_0.jpg\"\n",
    "    src_gt = src_dir / f\"{prefix}_img_gt.jpg\"\n",
    "    src_instr = src_dir / f\"{prefix}_instruction.txt\"\n",
    "\n",
    "    # 目标目录 /.../edit/<idx>/\n",
    "    dst_dir = OUT_ROOT / str(dst_idx)\n",
    "    dst_dir.mkdir(parents=True, exist_ok=True)\n",
    "\n",
    "    # 拷贝并重命名\n",
    "    mapping = [\n",
    "        (src_0,      dst_dir / \"orig_000.jpg\"),\n",
    "        (src_gt,     dst_dir / \"gt.jpg\"),\n",
    "        (src_instr,  dst_dir / \"instruction_token.txt\"),\n",
    "        # (src_generated, dst_dir / \"generated.jpg\"),\n",
    "    ]\n",
    "\n",
    "    for src, dst in mapping:\n",
    "        if not src.is_file():\n",
    "            print(f\"[WARN] Missing source file: {src}\")\n",
    "            continue\n",
    "        shutil.copy2(src, dst)\n",
    "        print(f\"[OK] {src} -> {dst}\")\n",
    "\n",
    "\n",
    "def main():\n",
    "    for idx, path in enumerate(GENERATED_PATHS, start=1):\n",
    "        print(f\"=== Processing sample #{idx} ===\")\n",
    "        copy_one_sample(path, idx)\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano_customized/test_sample112_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/1.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample1_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/2.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample6_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/3.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample22_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/4.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_background/test_sample0_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/5.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample3_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/6.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample5_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/7.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_Times_Change/test_sample18_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/8.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample11_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/9.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample71_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/10.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample73_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/11.jpg\n",
      "[OK] /fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample80_img_generated.jpg -> /fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit/12.jpg\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import shutil\n",
    "from pathlib import Path\n",
    "\n",
    "# ===== 按顺序给出的 12 个 generated 路径 =====\n",
    "GENERATED_PATHS = [\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano_customized/test_sample112_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample1_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample6_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_action/test_sample22_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_background/test_sample0_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample3_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_hairstyle/test_sample5_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_nano150k_Times_Change/test_sample18_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample11_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample71_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample73_img_generated.jpg\",\n",
    "    \"/fi-lib/workspace/zxy/code/Bagel_lora/results/all_special_select_lr00002_interleave_specialtoken_cosine_nolora_new/checkpoints/0050000/inference/test_x2edit/test_sample80_img_generated.jpg\",\n",
    "]\n",
    "\n",
    "# 目标根目录\n",
    "OUT_ROOT = Path(\"/fi-lib/workspace/zxy/data/Results/SIGMA_5/A_benchmark/edit\")\n",
    "OUT_ROOT.mkdir(parents=True, exist_ok=True)\n",
    "\n",
    "\n",
    "def main():\n",
    "    for idx, src in enumerate(GENERATED_PATHS, start=1):\n",
    "        src_path = Path(src)\n",
    "        if not src_path.is_file():\n",
    "            print(f\"[WARN] Missing file: {src_path}\")\n",
    "            continue\n",
    "\n",
    "        dst_path = OUT_ROOT / f\"{idx}.jpg\"\n",
    "        shutil.copy2(src_path, dst_path)\n",
    "        print(f\"[OK] {src_path} -> {dst_path}\")\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/1/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/10/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/11/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/12/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/2/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/3/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/4/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/5/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/6/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/7/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/8/instruction.txt\n",
      "[OK] Cleaned instruction saved: /fi-lib/workspace/zxy/data/A_benchmark/edit/9/instruction.txt\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "from pathlib import Path\n",
    "import re\n",
    "\n",
    "ROOT = Path(\"/fi-lib/workspace/zxy/data/A_benchmark/edit\")\n",
    "\n",
    "# 要删掉的所有 tokens（按你给的完整列表）\n",
    "TOKENS = [\n",
    "    \"<style>\", \"<subject>\", \"<id>\", \"<icon>\", \"<text>\", \"<pose>\", \"<layout>\", \n",
    "    \"<perspective>\", \"<lighting>\", \"<background>\", \"<environment>\", \n",
    "    \"<emotion>\", \"<clothing>\", \"<texture>\"\n",
    "]\n",
    "\n",
    "# 构造正则表达式（自动替换所有 token）\n",
    "pattern = re.compile(\"|\".join(map(re.escape, TOKENS)))\n",
    "\n",
    "\n",
    "def process_sample(sample_dir: Path):\n",
    "    src = sample_dir / \"instruction_token.txt\"\n",
    "    dst = sample_dir / \"instruction.txt\"\n",
    "\n",
    "    if not src.is_file():\n",
    "        print(f\"[WARN] No instruction_token.txt in {sample_dir}\")\n",
    "        return\n",
    "\n",
    "    # 读取内容\n",
    "    with open(src, \"r\", encoding=\"utf-8\") as f:\n",
    "        content = f.read()\n",
    "\n",
    "    # 删除 token\n",
    "    cleaned = pattern.sub(\"\", content)\n",
    "\n",
    "    # 写入新的 instruction.txt\n",
    "    with open(dst, \"w\", encoding=\"utf-8\") as f:\n",
    "        f.write(cleaned)\n",
    "\n",
    "    print(f\"[OK] Cleaned instruction saved: {dst}\")\n",
    "\n",
    "\n",
    "def main():\n",
    "    for sample in sorted(ROOT.iterdir()):\n",
    "        if sample.is_dir():\n",
    "            process_sample(sample)\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[OK] JSON saved to /fi-lib/workspace/zxy/data/A_benchmark/our_json/sigma_benchmark_json/edit.json\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import json\n",
    "from pathlib import Path\n",
    "\n",
    "EDIT_ROOT = Path(\"/fi-lib/workspace/zxy/data/A_benchmark/edit\")\n",
    "OUT_JSON = Path(\"/fi-lib/workspace/zxy/data/A_benchmark/our_json/sigma_benchmark_json/edit.json\")\n",
    "\n",
    "def main():\n",
    "    all_entries = []\n",
    "\n",
    "    for sample_dir in sorted(EDIT_ROOT.iterdir()):\n",
    "        if not sample_dir.is_dir():\n",
    "            continue\n",
    "\n",
    "        # original_images 可能是 jpg 或 png，可能有多个（orig_000, orig_001...）\n",
    "        original_images = sorted(\n",
    "            [str(p) for p in sample_dir.glob(\"orig_*.jpg\")] +\n",
    "            [str(p) for p in sample_dir.glob(\"orig_*.png\")]\n",
    "        )\n",
    "\n",
    "        if len(original_images) == 0:\n",
    "            print(f\"[WARN] No original image in {sample_dir}\")\n",
    "            continue\n",
    "\n",
    "        # generated image = gt.jpg\n",
    "        generated_image = sample_dir / \"gt.jpg\"\n",
    "        if not generated_image.exists():\n",
    "            print(f\"[WARN] Missing gt.jpg in {sample_dir}\")\n",
    "            continue\n",
    "\n",
    "        # instruction file\n",
    "        instruction_file = sample_dir / \"instruction.txt\"\n",
    "        if not instruction_file.exists():\n",
    "            print(f\"[WARN] Missing instruction.txt in {sample_dir}\")\n",
    "            continue\n",
    "\n",
    "        entry = {\n",
    "            \"original_images\": original_images,\n",
    "            \"generated_image\": str(generated_image),\n",
    "            \"prompt_used\": str(instruction_file),\n",
    "            \"data_prep_method\": \"edit\"\n",
    "        }\n",
    "\n",
    "        all_entries.append(entry)\n",
    "\n",
    "    # 创建输出目录\n",
    "    OUT_JSON.parent.mkdir(parents=True, exist_ok=True)\n",
    "\n",
    "    # 写 json（美化格式）\n",
    "    with open(OUT_JSON, \"w\", encoding=\"utf-8\") as f:\n",
    "        json.dump(all_entries, f, indent=4, ensure_ascii=False)\n",
    "\n",
    "    print(f\"[OK] JSON saved to {OUT_JSON}\")\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "zxy_env",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}