add face attribute labels to README and implement category recording function

README.md

@@ -2,6 +2,26 @@
 license: mit
 ---
 
+## Labels
+
+```txt
+•	smile：smiling, no smile
+•	pitch：downward, neutral, upward
+•	roll：tilted, neutral
+•	yaw：turned left, frontal, turned right
+•	gender：male, female
+•	age：adult, baby, child, teenager, senior
+•	facialHair：moustache, beard, sideburns, none
+•	glasses：NoGlasses, SwimmingGoggles, Sunglasses, ReadingGlasses
+•	emotion：happiness, sadness, surprise, contempt, disgust, neutral, anger, fear
+•	blur：high, medium, low
+•	exposure：goodExposure, underExposure, overExposure
+•	noise：high, medium, low
+•	makeup：eye makeup, lip makeup, no makeup
+•	accessories：glasses, mask, headwear, none
+•	occlusion：foreheadOccluded, mouthOccluded, eyeOccluded, none
+•	hair：other, blond, gray, bald, unknown, red, brown, black
+```
 
 ## Acknowledgements
 

main/main.py

@@ -1,6 +1,5 @@
 import json
 import os
-import sys
 
 def parse_face_attributes(json_file):
     with open(json_file, 'r', encoding='utf-8') as file:
@@ -13,18 +12,17 @@ def parse_face_attributes(json_file):
     
     result = {}
     
-    # 1. 處理笑容：根據 smile 值轉換為類別
-    # 若 smile 大於 0.1 則視為「smiling」，否則視為「no smile」
+    # 1. 笑容：若 smile 大於 0.1 則視為 "smiling"，否則 "no smile"
     smile_value = face_attr.get("smile", 0)
     result['smile'] = "smiling" if smile_value > 0.1 else "no smile"
     
-    # 2. 處理頭部姿勢 (headPose)
+    # 2. 頭部姿勢 (headPose)
     head_pose = face_attr.get("headPose", {})
     pitch = head_pose.get("pitch", 0)
     roll = head_pose.get("roll", 0)
     yaw = head_pose.get("yaw", 0)
     
-    # pitch：若大於 5 則「upward」，小於 -5 則「downward」，否則「neutral」
+    # pitch：大於 5 為 "upward"，小於 -5 為 "downward"，否則 "neutral"
     if pitch > 5:
         pitch_cat = "upward"
     elif pitch < -5:
@@ -32,10 +30,10 @@ def parse_face_attributes(json_file):
     else:
         pitch_cat = "neutral"
     
-    # roll：若絕對值大於 5 則「tilted」，否則「neutral」
+    # roll：絕對值大於 5 為 "tilted"，否則 "neutral"
     roll_cat = "tilted" if abs(roll) > 5 else "neutral"
     
-    # yaw：若大於 15 則「turned right」，小於 -15 則「turned left」，否則「frontal」
+    # yaw：大於 15 為 "turned right"，小於 -15 為 "turned left"，否則 "frontal"
     if yaw > 15:
         yaw_cat = "turned right"
     elif yaw < -15:
@@ -45,11 +43,10 @@ def parse_face_attributes(json_file):
     
     result['headPose'] = {"pitch": pitch_cat, "roll": roll_cat, "yaw": yaw_cat}
     
-    # 3. 性別 (gender) 為類別型資料，直接取用
+    # 3. 性別 (gender)
     result["gender"] = face_attr.get("gender", "unknown")
     
-    # 4. 年齡 (age)：將數值轉換為類別
-    # 定義閾值：小於 2 歲視為 "baby"，2-10 歲視為 "child"，10-20 歲視為 "teenager"，20-60 歲視為 "adult"，60 歲以上視為 "senior"
+    # 4. 年齡 (age)：依年齡數值分類
     age = face_attr.get("age", 0)
     if age < 2:
         age_cat = "baby"
@@ -63,18 +60,18 @@ def parse_face_attributes(json_file):
         age_cat = "senior"
     result["age"] = age_cat
     
-    # 5. 面部毛髮 (facialHair)：檢查各項是否大於 0.1，若皆不超過則視為 "none"
+    # 5. 面部毛髮 (facialHair)：若各項均不超過 0.1 則為 "none"
     facial_hair = face_attr.get("facialHair", {})
     hair_types = []
     for key, value in facial_hair.items():
-        if value > 0.1:  # 可調整閾值
+        if value > 0.1:
             hair_types.append(key)
     result["facialHair"] = "none" if not hair_types else ", ".join(hair_types)
     
-    # 6. 眼鏡狀態 (glasses) 直接取用原始資料
+    # 6. 眼鏡狀態 (glasses)
     result["glasses"] = face_attr.get("glasses", "NoGlasses")
     
-    # 7. 情緒 (emotion)：取出分數最高的情緒作為類別
+    # 7. 情緒 (emotion)：取分數最高的情緒
     emotion = face_attr.get("emotion", {})
     if emotion:
         emotion_category = max(emotion, key=emotion.get)
@@ -82,15 +79,15 @@ def parse_face_attributes(json_file):
         emotion_category = "unknown"
     result["emotion"] = emotion_category
     
-    # 8. 模糊度 (blur)：採用 blurLevel 作為類別
+    # 8. 模糊度 (blur)：以 blurLevel 為類別
     blur = face_attr.get("blur", {})
     result["blur"] = blur.get("blurLevel", "unknown")
     
-    # 9. 曝光 (exposure)：採用 exposureLevel 作為類別
+    # 9. 曝光 (exposure)：以 exposureLevel 為類別
     exposure = face_attr.get("exposure", {})
     result["exposure"] = exposure.get("exposureLevel", "unknown")
     
-    # 10. 噪音 (noise)：採用 noiseLevel 作為類別
+    # 10. 噪音 (noise)：以 noiseLevel 為類別
     noise = face_attr.get("noise", {})
     result["noise"] = noise.get("noiseLevel", "unknown")
     
@@ -103,16 +100,16 @@ def parse_face_attributes(json_file):
         makeup_categories.append("lip makeup")
     result["makeup"] = "no makeup" if not makeup_categories else ", ".join(makeup_categories)
     
-    # 12. 配件 (accessories)：若空則回傳 "none"
+    # 12. 配件 (accessories)：若無配件則回傳 "none"
     accessories = face_attr.get("accessories", [])
     result["accessories"] = "none" if not accessories else ", ".join([acc.get("type", "unknown") for acc in accessories])
     
-    # 13. 遮擋 (occlusion)：若皆為 False 則 "none"，否則列出被遮擋的部位
+    # 13. 遮擋 (occlusion)：若皆為 False 則為 "none"，否則列出被遮擋的部位
     occlusion = face_attr.get("occlusion", {})
     occlusion_list = [k for k, v in occlusion.items() if v]
     result["occlusion"] = "none" if not occlusion_list else ", ".join(occlusion_list)
     
-    # 14. 頭髮 (hair)：若 bald 大於 0.5 則視為 bald，否則取 hairColor 中信心最高的顏色
+    # 14. 頭髮 (hair)：若 bald 大於 0.5 則為 "bald"，否則取 hairColor 中信心最高的顏色
     hair = face_attr.get("hair", {})
     if hair:
         if hair.get("bald", 0) > 0.5:
@@ -130,29 +127,47 @@ def parse_face_attributes(json_file):
     
     return result
 
-def test_json(file_path):
-    with open(file_path, 'r', encoding='utf-8') as file:
-        data = json.load(file)
-    return len(data)
+def record_all_categories(directory):
+    """
+    讀取指定目錄下所有 JSON 檔案，並統計每個屬性出現的所有類別
+    """
+    categories = {}  # 用來儲存各屬性的所有類別 (以 set 方式儲存避免重複)
+    
+    # 遍歷資料夾中所有 JSON 檔案
+    error_cnt = 0
+    for filename in os.listdir(directory):
+        if not filename.endswith(".json"):
+            continue
+        file_path = os.path.join(directory, filename)
+        try:
+            parsed_result = parse_face_attributes(file_path)
+        except Exception as e:
+            error_cnt += 1
+            # print(f"處理檔案 {filename} 時發生錯誤: {e}")
+            continue
+        
+        # 將每個屬性對應的類別加入集合中
+        for key, value in parsed_result.items():
+            # 若屬性值為字典（例如 headPose），則進一步處理各子項
+            if isinstance(value, dict):
+                for sub_key, sub_val in value.items():
+                    if sub_key not in categories:
+                        categories[sub_key] = set()
+                    categories[sub_key].add(sub_val)
+            else:
+                if key not in categories:
+                    categories[key] = set()
+                categories[key].add(value)
+    print(f"總共處理 {len(os.listdir(directory))} 個檔案，其中 {error_cnt} 個檔案發生錯誤。")
+    return categories
 
-# 範例使用方式：
 if __name__ == "__main__":
-    from pprint import pprint
-    # 假設 JSON 檔案名稱為 "face_data.json"
-    parsed_data = parse_face_attributes("/root/siglip-recursion-ffhq-thumbnails/main/temp_data/ffhq-features-dataset/json/01111.json")
-    pprint(parsed_data)
-    # file_list = os.listdir("/root/siglip-recursion-ffhq-thumbnails/main/temp_data/ffhq-features-dataset/json")
-    # cnt = {}
-    # outlier = {}
-    # for file in file_list:
-    #     file_path = os.path.join("/root/siglip-recursion-ffhq-thumbnails/main/temp_data/ffhq-features-dataset/json", file)
-    #     length = test_json(file_path)
-    #     if length in cnt:
-    #         cnt[length] += 1
-    #     else:
-    #         cnt[length] = 1
-    #     if length > 1:
-    #         outlier[file] = length
-    # print(outlier)
-    # print(cnt)
-    # print(f"total: {sum(cnt.values())}")
+    # 設定 JSON 檔案所在目錄（請根據實際路徑修改）
+    directory = "/root/siglip-recursion-ffhq-thumbnails/main/temp_data/ffhq-features-dataset/json"
+    
+    # 紀錄所有類別
+    all_categories = record_all_categories(directory)
+    
+    # 印出各屬性出現的所有類別
+    for attribute, values in all_categories.items():
+        print(f"{attribute}: {', '.join(values)}")