Agent/examples/find knowledge/generate_mappings.py

"""
生成每个特征维度的 mapping.json 文件
记录特征与制作表的对应关系
"""

import os
import json

WORKDIR = os.path.dirname(os.path.abspath(__file__))
OUTPUT_DIR = os.path.join(WORKDIR, 'output', 'features')

# 读取所有制作表数据
all_data = {}
for i in range(1, 10):
    with open(os.path.join(WORKDIR, 'input', f'写生油画__img_{i}_合并评分.json'), 'r', encoding='utf-8') as f:
        all_data[f'img_{i}'] = json.load(f)

# 读取亮点数据
with open(os.path.join(WORKDIR, 'input', '写生油画__post_highlight_简化版.json'), 'r', encoding='utf-8') as f:
    highlights = json.load(f)

# ============================================================
# 1. OpenPose 骨架 - mapping
# ============================================================
openpose_mapping = {
    "dimension": "openpose_skeleton",
    "description": "人体姿态骨架图，使用OpenPose提取关节点坐标，捕捉人物的站姿、跪姿、侧身等姿态",
    "tool": "controlnet_aux.OpenposeDetector (lllyasviel/ControlNet)",
    "format": "PNG (黑底彩色骨架图)",
    "highlight_clusters": ["cluster_1 (优雅的白裙写生少女)", "cluster_6 (引导视线的构图技巧)"],
    "mappings": []
}

# 每张图片的骨架对应关系
pose_info = {
    "img_1": {"段落": "段落1.1", "段落名称": "人物", "姿态描述": "侧身背对镜头，站立作画，右手持笔，左手持调色板"},
    "img_2": {"段落": "段落2.1", "段落名称": "人物", "姿态描述": "背对镜头，站立作画，逆光轮廓"},
    "img_3": {"段落": "段落3.1", "段落名称": "人物", "姿态描述": "背对镜头，跪坐在草地上作画"},
    "img_4": {"段落": "段落4.1", "段落名称": "人物", "姿态描述": "侧身面对镜头，站立，右手持笔上举，左手持调色板"},
    "img_5": {"段落": "段落5.1", "段落名称": "人物", "姿态描述": "近景，上半身，左手持调色板特写"},
    "img_6": {"段落": "段落6.1", "段落名称": "人物", "姿态描述": "侧身背对，过肩视角，右手持笔作画"},
    "img_7": {"段落": "段落7.1", "段落名称": "人物", "姿态描述": "侧脸特写，双手捧玫瑰花，闭眼闻花"},
    "img_8": {"段落": "段落8.1", "段落名称": "人物", "姿态描述": "侧身面对镜头，站立作画"},
    "img_9": {"段落": "段落9.1", "段落名称": "人物", "姿态描述": "背对镜头，远景，全身站立作画"},
}

for img_name, info in pose_info.items():
    # 找到对应段落的评分
    img_data = all_data[img_name][0]
    
    # 找到人物段落的形式评分
    person_score = 0
    for sub in img_data.get('子段落', []):
        if '人物' in sub.get('名称', ''):
            person_score = sub.get('形式', {}).get('评分详情', {}).get('combined_score', 0)
            break
    
    openpose_mapping["mappings"].append({
        "图片": img_name,
        "特征文件": f"{img_name}.png",
        "段落ID": info["段落"],
        "段落名称": info["段落名称"],
        "维度": "实质",
        "特征路径": f"子段落.{info['段落']}.形式",
        "具体特征": "拍摄角度 / 构图",
        "姿态描述": info["姿态描述"],
        "亮点关联": "cluster_1 (优雅的白裙写生少女) / cluster_6 (引导视线的构图技巧)",
        "参考评分": person_score
    })

with open(os.path.join(OUTPUT_DIR, 'openpose_skeleton', 'mapping.json'), 'w', encoding='utf-8') as f:
    json.dump(openpose_mapping, f, indent=2, ensure_ascii=False)
print("openpose_skeleton/mapping.json 生成完成")

# ============================================================
# 2. Depth Map - mapping
# ============================================================
depth_mapping = {
    "dimension": "depth_map",
    "description": "深度图，使用MiDaS提取场景空间深度信息，捕捉前景人物/画架与背景草地/树木的层次关系",
    "tool": "controlnet_aux.MidasDetector (lllyasviel/Annotators)",
    "format": "PNG (灰度深度图，亮=近，暗=远)",
    "highlight_clusters": ["cluster_4 (唯美梦幻的光影与景深)", "cluster_3 (清新雅致的白绿配色)"],
    "mappings": []
}

depth_info = {
    "img_1": {"段落": "段落1", "段落名称": "户外绘画场景", "景深特征": "中景，人物/画架在前景，绿色树木在远景，轻微虚化"},
    "img_2": {"段落": "段落2", "段落名称": "户外绘画场景", "景深特征": "逆光，背景强烈虚化，人物轮廓光，梦幻散景"},
    "img_3": {"段落": "段落3", "段落名称": "户外绘画场景", "景深特征": "跪姿，人物低于正常视角，背景建筑远景"},
    "img_4": {"段落": "段落4", "段落名称": "户外绘画场景", "景深特征": "侧前方视角，人物与画架同一景深平面"},
    "img_5": {"段落": "段落5", "段落名称": "户外绘画场景", "景深特征": "近景特写，调色板在最近景，背景极度虚化"},
    "img_6": {"段落": "段落6", "段落名称": "户外绘画场景", "景深特征": "特写，极浅景深，背景完全虚化"},
    "img_7": {"段落": "段落7", "段落名称": "人物与玫瑰花", "景深特征": "侧脸特写，玫瑰花在最近景，背景草地虚化"},
    "img_8": {"段落": "段落8", "段落名称": "户外绘画场景", "景深特征": "中景，人物与画架在前景，背景树木虚化"},
    "img_9": {"段落": "段落9", "段落名称": "户外绘画场景", "景深特征": "远景，人物较小，背景建筑和树木可见"},
}

for img_name, info in depth_info.items():
    depth_mapping["mappings"].append({
        "图片": img_name,
        "特征文件": f"{img_name}.png",
        "段落ID": info["段落"],
        "段落名称": info["段落名称"],
        "维度": "形式",
        "特征路径": f"形式.清晰度 / 段落关系.段内关系.景深",
        "具体特征": "清晰度 / 景深",
        "景深描述": info["景深特征"],
        "亮点关联": "cluster_4 (唯美梦幻的光影与景深)",
        "参考评分": all_data[img_name][0].get('形式', {}).get('清晰度', {}).get('评分详情', {}).get('combined_score', 0)
    })

with open(os.path.join(OUTPUT_DIR, 'depth_map', 'mapping.json'), 'w', encoding='utf-8') as f:
    json.dump(depth_mapping, f, indent=2, ensure_ascii=False)
print("depth_map/mapping.json 生成完成")

# ============================================================
# 3. Lineart Edge - mapping
# ============================================================
lineart_mapping = {
    "dimension": "lineart_edge",
    "description": "线稿/边缘图，使用Lineart检测器提取图像轮廓线条，捕捉人物轮廓、服装褶皱、画架结构等",
    "tool": "controlnet_aux.LineartDetector (lllyasviel/Annotators)",
    "format": "PNG (白底黑线线稿图)",
    "highlight_clusters": ["cluster_1 (优雅的白裙写生少女)", "cluster_2_props (构建叙事的写生道具)"],
    "mappings": []
}

lineart_info = {
    "img_1": {
        "段落": "段落1.1.2", "段落名称": "身体", "维度": "形式",
        "特征": "服装款式", "描述": "白色长裙轮廓线，袖子宽松，腰部收紧，裙摆飘逸，背部系带细节"
    },
    "img_2": {
        "段落": "段落2.1.2", "段落名称": "身体", "维度": "形式",
        "特征": "服装款式", "描述": "白色长裙轮廓，V字露背设计，A字裙摆，逆光轮廓光"
    },
    "img_3": {
        "段落": "段落3.1.2", "段落名称": "身体", "维度": "形式",
        "特征": "服装款式", "描述": "跪坐姿态下白裙轮廓，裙摆自然垂坠，背部V字系带"
    },
    "img_4": {
        "段落": "段落4.2", "段落名称": "画架", "维度": "形式",
        "特征": "构图", "描述": "画架三脚架结构线条，画布矩形轮廓，人物与画架的空间线条关系"
    },
    "img_5": {
        "段落": "段落5.1.3", "段落名称": "调色板", "维度": "实质",
        "特征": "颜色", "描述": "调色板轮廓，颜料堆积纹理线条，手部握持姿态"
    },
    "img_6": {
        "段落": "段落6.2.1", "段落名称": "画布", "维度": "实质",
        "特征": "笔触", "描述": "画布上油画笔触线条，颜料堆叠纹理，画框轮廓"
    },
    "img_7": {
        "段落": "段落7.2.1", "段落名称": "花朵", "维度": "实质",
        "特征": "清晰度", "描述": "玫瑰花瓣层叠轮廓，花瓣边缘线条，茎叶结构"
    },
    "img_8": {
        "段落": "段落8.1.2", "段落名称": "身体", "维度": "形式",
        "特征": "服装款式", "描述": "白色长裙轮廓，侧身姿态线条，手持画笔动作"
    },
    "img_9": {
        "段落": "段落9.1.2", "段落名称": "身体", "维度": "形式",
        "特征": "服装款式", "描述": "远景全身白裙轮廓，露背设计，裙摆及地"
    },
}

for img_name, info in lineart_info.items():
    lineart_mapping["mappings"].append({
        "图片": img_name,
        "特征文件": f"{img_name}.png",
        "段落ID": info["段落"],
        "段落名称": info["段落名称"],
        "维度": info["维度"],
        "特征路径": f"子段落.{info['段落']}.形式.{info['特征']}",
        "具体特征": info["特征"],
        "线稿描述": info["描述"],
        "亮点关联": "cluster_1 (优雅的白裙写生少女) / cluster_2_props (写生道具)"
    })

with open(os.path.join(OUTPUT_DIR, 'lineart_edge', 'mapping.json'), 'w', encoding='utf-8') as f:
    json.dump(lineart_mapping, f, indent=2, ensure_ascii=False)
print("lineart_edge/mapping.json 生成完成")

# ============================================================
# 4. Color Palette - mapping
# ============================================================
color_palette_mapping = {
    "dimension": "color_palette",
    "description": "主色调调色板，使用ColorThief提取图片的8个主要颜色，捕捉白绿配色、调色板颜料色彩等",
    "tool": "colorthief (Python库)",
    "format": "PNG (色块可视化) + JSON (颜色数值)",
    "highlight_clusters": ["cluster_3 (清新雅致的白绿配色)", "cluster_2_texture (斑斓厚重的油画颜料)"],
    "mappings": []
}

palette_info = {
    "img_1": {
        "段落": "段落1", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "色彩饱和度", "描述": "主色：纯白(白裙)、草绿(背景)、深棕(调色板)、多彩颜料色"
    },
    "img_2": {
        "段落": "段落2", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "色彩饱和度", "描述": "逆光色调：暖黄光晕、白色、绿色，整体偏暖"
    },
    "img_3": {
        "段落": "段落3", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "色彩饱和度", "描述": "白绿配色，远景建筑灰色，整体清新"
    },
    "img_4": {
        "段落": "段落4", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "色彩饱和度", "描述": "明亮自然光，白裙纯白，绿色鲜艳，调色板多彩"
    },
    "img_5": {
        "段落": "段落5.1.3", "段落名称": "调色板", "维度": "实质",
        "特征": "颜色", "描述": "调色板颜料色：深绿、浅绿、蓝、红、黄、白、紫、黑等10+种颜色"
    },
    "img_6": {
        "段落": "段落6.2.1", "段落名称": "画布", "维度": "实质",
        "特征": "色彩", "描述": "画布油画色：绿色、蓝色为主，夹杂白、黄、紫、棕，颜料堆叠"
    },
    "img_7": {
        "段落": "段落7", "段落名称": "人物与玫瑰花", "维度": "形式",
        "特征": "色彩饱和度", "描述": "白玫瑰纯白、肤色米白、绿色背景，整体清新淡雅"
    },
    "img_8": {
        "段落": "段落8", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "色彩饱和度", "描述": "白绿配色，光线均匀，整体明亮清新"
    },
    "img_9": {
        "段落": "段落9", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "色彩饱和度", "描述": "远景色调，绿色鲜艳，白裙洁净，远处建筑浅灰"
    },
}

for img_name, info in palette_info.items():
    color_palette_mapping["mappings"].append({
        "图片": img_name,
        "特征文件_PNG": f"{img_name}.png",
        "特征文件_JSON": f"{img_name}.json",
        "段落ID": info["段落"],
        "段落名称": info["段落名称"],
        "维度": info["维度"],
        "特征路径": f"形式.{info['特征']}",
        "具体特征": info["特征"],
        "色彩描述": info["描述"],
        "亮点关联": "cluster_3 (清新雅致的白绿配色) / cluster_2_texture (斑斓厚重的油画颜料)"
    })

with open(os.path.join(OUTPUT_DIR, 'color_palette', 'mapping.json'), 'w', encoding='utf-8') as f:
    json.dump(color_palette_mapping, f, indent=2, ensure_ascii=False)
print("color_palette/mapping.json 生成完成")

# ============================================================
# 5. Bokeh Mask - mapping
# ============================================================
bokeh_mapping = {
    "dimension": "bokeh_mask",
    "description": "景深虚化遮罩，基于深度图和清晰度分析推导，捕捉大光圈浅景深效果，区分清晰主体与虚化背景",
    "tool": "自定义算法 (MiDaS深度图 + 拉普拉斯清晰度)",
    "format": "PNG (灰度遮罩，亮=清晰主体，暗=虚化背景)",
    "highlight_clusters": ["cluster_4 (唯美梦幻的光影与景深)"],
    "mappings": []
}

bokeh_info = {
    "img_1": {"段落": "段落1.3", "段落名称": "背景", "维度": "形式", "特征": "清晰度", "描述": "背景树木轻微虚化，人物/画架清晰"},
    "img_2": {"段落": "段落2.3", "段落名称": "背景", "维度": "形式", "特征": "光照", "描述": "逆光强烈虚化，背景散景(Bokeh)效果明显，光斑圆形"},
    "img_3": {"段落": "段落3.3", "段落名称": "背景", "维度": "形式", "特征": "清晰度", "描述": "背景建筑和树木虚化，人物清晰"},
    "img_4": {"段落": "段落4.3", "段落名称": "背景", "维度": "形式", "特征": "清晰度", "描述": "背景树木虚化，人物与画架清晰"},
    "img_5": {"段落": "段落5.3", "段落名称": "背景", "维度": "形式", "特征": "清晰度", "描述": "背景极度虚化，调色板和手部清晰"},
    "img_6": {"段落": "段落6.3", "段落名称": "背景", "维度": "形式", "特征": "清晰度", "描述": "背景完全虚化，极浅景深，人物和画布清晰"},
    "img_7": {"段落": "段落7.3", "段落名称": "背景", "维度": "形式", "特征": "清晰度", "描述": "背景草地虚化，人物侧脸和玫瑰花清晰"},
    "img_8": {"段落": "段落8.3", "段落名称": "背景", "维度": "形式", "特征": "清晰度", "描述": "背景树木虚化，人物和画架清晰"},
    "img_9": {"段落": "段落9.3", "段落名称": "背景", "维度": "形式", "特征": "清晰度", "描述": "远景，整体清晰度较高，背景建筑可见"},
}

for img_name, info in bokeh_info.items():
    bokeh_mapping["mappings"].append({
        "图片": img_name,
        "特征文件": f"{img_name}.png",
        "段落ID": info["段落"],
        "段落名称": info["段落名称"],
        "维度": info["维度"],
        "特征路径": f"子段落.{info['段落']}.形式.{info['特征']}",
        "具体特征": info["特征"],
        "景深描述": info["描述"],
        "亮点关联": "cluster_4 (唯美梦幻的光影与景深)"
    })

with open(os.path.join(OUTPUT_DIR, 'bokeh_mask', 'mapping.json'), 'w', encoding='utf-8') as f:
    json.dump(bokeh_mapping, f, indent=2, ensure_ascii=False)
print("bokeh_mask/mapping.json 生成完成")

# ============================================================
# 6. Semantic Segmentation - mapping
# ============================================================
seg_mapping = {
    "dimension": "semantic_segmentation",
    "description": "语义分割图，基于颜色聚类将图像分割为主要语义区域：白裙人物/绿色背景/调色板/画架/画布/皮肤",
    "tool": "sklearn.KMeans + OpenCV (LAB颜色空间聚类)",
    "format": "PNG (彩色分割图，不同颜色代表不同语义区域)",
    "highlight_clusters": ["cluster_1 (优雅的白裙写生少女)", "cluster_3 (清新雅致的白绿配色)", "cluster_5 (虚实呼应的画中画结构)"],
    "mappings": []
}

seg_info = {
    "img_1": {
        "段落": "段落1", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "构图", "描述": "分割区域：白裙人物(右侧60%)、绿色背景(左上40%)、画架(左侧)、画布(中央)"
    },
    "img_2": {
        "段落": "段落2", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "构图", "描述": "分割区域：人物背影(中央)、逆光背景(上方亮区)、绿色草地(下方)"
    },
    "img_3": {
        "段落": "段落3", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "构图", "描述": "分割区域：跪坐人物(右侧)、画架(中央)、草地(前景)、树木+建筑(背景)"
    },
    "img_4": {
        "段落": "段落4", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "构图", "描述": "分割区域：侧身人物(右侧)、空白画布(左侧)、绿色背景(上方)"
    },
    "img_5": {
        "段落": "段落5.1.3", "段落名称": "调色板", "维度": "实质",
        "特征": "颜色", "描述": "分割区域：白色服装(大面积)、多彩调色板(中央)、绿色背景(下方)"
    },
    "img_6": {
        "段落": "段落6.2.1", "段落名称": "画布", "维度": "实质",
        "特征": "内容主题", "描述": "分割区域：人物背部(右侧)、油画画布(中央左侧)、虚化绿色背景"
    },
    "img_7": {
        "段落": "段落7", "段落名称": "人物与玫瑰花", "维度": "实质",
        "特征": "花朵颜色", "描述": "分割区域：人物侧脸(右侧)、白色玫瑰(中央)、绿色虚化背景"
    },
    "img_8": {
        "段落": "段落8", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "构图", "描述": "分割区域：侧身人物(右侧)、空白画布(左侧)、绿色背景(上方)"
    },
    "img_9": {
        "段落": "段落9", "段落名称": "户外绘画场景", "维度": "形式",
        "特征": "构图", "描述": "分割区域：远景人物(右侧小)、画架(右侧)、大面积草地(左侧)、树木+建筑(背景)"
    },
}

for img_name, info in seg_info.items():
    seg_mapping["mappings"].append({
        "图片": img_name,
        "特征文件": f"{img_name}.png",
        "段落ID": info["段落"],
        "段落名称": info["段落名称"],
        "维度": info["维度"],
        "特征路径": f"形式.{info['特征']}",
        "具体特征": info["特征"],
        "分割描述": info["描述"],
        "亮点关联": "cluster_1 / cluster_3 / cluster_5"
    })

with open(os.path.join(OUTPUT_DIR, 'semantic_segmentation', 'mapping.json'), 'w', encoding='utf-8') as f:
    json.dump(seg_mapping, f, indent=2, ensure_ascii=False)
print("semantic_segmentation/mapping.json 生成完成")

# ============================================================
# 7. Color Distribution - mapping
# ============================================================
dist_mapping = {
    "dimension": "color_distribution",
    "description": "HSV色彩分布向量，包含色相/饱和度/明度直方图及统计特征，量化白绿配色比例、整体亮度等",
    "tool": "OpenCV calcHist (HSV颜色空间)",
    "format": "JSON (数值向量) + PNG (色相直方图可视化)",
    "highlight_clusters": ["cluster_3 (清新雅致的白绿配色)", "cluster_4 (唯美梦幻的光影与景深)"],
    "mappings": []
}

dist_info = {
    "img_1": {"段落": "段落1", "特征": "色彩饱和度", "描述": "白色比例高(白裙)，绿色比例高(背景)，中等亮度"},
    "img_2": {"段落": "段落2", "特征": "光照", "描述": "逆光高亮，整体亮度偏高，暖黄色调"},
    "img_3": {"段落": "段落3", "特征": "色彩饱和度", "描述": "白绿配色，整体清新，远景建筑降低饱和度"},
    "img_4": {"段落": "段落4", "特征": "色彩饱和度", "描述": "明亮自然光，高饱和度绿色，白色纯净"},
    "img_5": {"段落": "段落5.1.3", "特征": "颜色种类", "描述": "调色板多色，绿色主导，白色服装大面积"},
    "img_6": {"段落": "段落6.2.1", "特征": "色彩", "描述": "画布绿蓝色调，背景虚化绿色，整体饱和度高"},
    "img_7": {"段落": "段落7", "特征": "色彩饱和度", "描述": "白色主导(玫瑰+服装)，绿色背景，整体淡雅"},
    "img_8": {"段落": "段落8", "特征": "色彩饱和度", "描述": "白绿配色，光线均匀，整体明亮"},
    "img_9": {"段落": "段落9", "特征": "色彩饱和度", "描述": "远景色调，绿色鲜艳，白色洁净，整体清新"},
}

for img_name, info in dist_info.items():
    dist_mapping["mappings"].append({
        "图片": img_name,
        "特征文件_JSON": f"{img_name}.json",
        "特征文件_PNG": f"{img_name}.png",
        "段落ID": info["段落"],
        "段落名称": "户外绘画场景",
        "维度": "形式",
        "特征路径": f"形式.{info['特征']}",
        "具体特征": info["特征"],
        "色彩描述": info["描述"],
        "亮点关联": "cluster_3 (清新雅致的白绿配色)"
    })

with open(os.path.join(OUTPUT_DIR, 'color_distribution', 'mapping.json'), 'w', encoding='utf-8') as f:
    json.dump(dist_mapping, f, indent=2, ensure_ascii=False)
print("color_distribution/mapping.json 生成完成")

print("\n=== 所有 mapping.json 生成完成 ===")