Blame for inference/explain_data/token_groups.json

{
"_meta": {
"tokens": {
"S": {
"name_zh": "主體 Subject",
"name_en": "Subject",
"question_zh": "拍什麼",
"n": 10
},
"C": {
"name_zh": "構圖 Composition",
"name_en": "Composition",
"question_zh": "怎麼擺",
"n": 22
},
"A": {
"name_zh": "美學 Aesthetics",
"name_en": "Aesthetics",
"question_zh": "美不美",
"n": 8
},
"I": {
"name_zh": "場景完整性 Integrity",
"name_en": "Integrity",
"question_zh": "切到誰",
"n": 20
}
},
"note": "feature -> token assignment with hand-polished zh/en phrasings."
},
"features": {
"vmix_max": {
"token": "S",
"zh": "視線最先落點的強度",
"en": "strength of where the eye lands first"
},
"v1_max": {
"token": "S",
"zh": "主視覺焦點的強度",
"en": "strength of the primary focal point"
},
"v1_cov": {
"token": "S",
"zh": "主體在畫面中的視覺佔比",
"en": "how much of the frame the subject commands"
},
"inspy_capture": {
"token": "S",
"zh": "主體收進畫面的完整度",
"en": "how fully the subject is kept in frame"
},
"inspy_max": {
"token": "S",
"zh": "主體最顯眼處的強度",
"en": "peak prominence of the subject"
},
"subject_size_ratio": {
"token": "S",
"zh": "主體的景別大小",
"en": "how large the subject sits in the frame"
},
"sam_mask_area_ratio": {
"token": "S",
"zh": "主體佔畫面的面積比",
"en": "the subject's share of the frame area"
},
"sam_mask_centered_x": {
"token": "S",
"zh": "主體的水平位置",
"en": "the subject's horizontal placement"
},
"sam_mask_centered_y": {
"token": "S",
"zh": "主體的垂直位置",
"en": "the subject's vertical placement"
},
"sam_mask_vs_inspy_iou": {
"token": "S",
"zh": "兩套主體判讀的一致程度",
"en": "agreement between two subject readings"
},
"thirds_align": {
"token": "C",
"zh": "三分法對齊程度",
"en": "rule-of-thirds alignment"
},
"sym_v_diff": {
"token": "C",
"zh": "左右對稱感",
"en": "left-right symmetry"
},
"sym_diag_diff": {
"token": "C",
"zh": "對角線對稱感",
"en": "diagonal symmetry"
},
"tri_top_apex": {
"token": "C",
"zh": "上聚頂的三角構圖",
"en": "upward-pointing triangular composition"
},
"tri_bot_apex": {
"token": "C",
"zh": "下聚底的三角構圖",
"en": "downward-pointing triangular composition"
},
"visual_weight_x": {
"token": "C",
"zh": "左右視覺重量的平衡",
"en": "left-right balance of visual weight"
},
"visual_weight_y": {
"token": "C",
"zh": "上下視覺重量的平衡",
"en": "top-bottom balance of visual weight"
},
"ca_dist": {
"token": "C",
"zh": "主體偏離畫面中心的幅度",
"en": "how far the subject sits from the frame center"
},
"area_ratio": {
"token": "C",
"zh": "取景的鬆緊",
"en": "how tight or loose the framing is"
},
"log_aspect_diff": {
"token": "C",
"zh": "長寬比與指定比例的吻合",
"en": "how well the aspect ratio matches the request"
},
"sam_thirds_q_mean": {
"token": "C",
"zh": "各物件落在三分線上的平均程度",
"en": "how consistently objects sit on the thirds lines"
},
"sam_thirds_q_max": {
"token": "C",
"zh": "最貼三分線的物件對齊度",
"en": "the strongest object-on-thirds alignment"
},
"sam_thirds_vs_center": {
"token": "C",
"zh": "三分擺位相對置中擺位的傾向",
"en": "lean toward thirds versus centered placement"
},
"clip_sim_pro": {
"token": "C",
"zh": "與『專業構圖』的語義貼近度",
"en": "closeness to a 'professional composition' cue"
},
"clip_sim_balanced": {
"token": "C",
"zh": "與『平衡構圖』的語義貼近度",
"en": "closeness to a 'balanced composition' cue"
},
"clip_sim_leading": {
"token": "C",
"zh": "與『引導線構圖』的語義貼近度",
"en": "closeness to a 'leading lines' cue"
},
"clip_sim_thirds": {
"token": "C",
"zh": "與『三分法構圖』的語義貼近度",
"en": "closeness to a 'rule of thirds' cue"
},
"clip_sim_symmetric": {
"token": "C",
"zh": "與『對稱構圖』的語義貼近度",
"en": "closeness to a 'symmetric composition' cue"
},
"clip_sim_negspace": {
"token": "C",
"zh": "與『留白構圖』的語義貼近度",
"en": "closeness to a 'negative space' cue"
},
"clip_sim_centered": {
"token": "C",
"zh": "與『置中構圖』的語義貼近度",
"en": "closeness to a 'centered composition' cue"
},
"clip_sim_dynamic": {
"token": "C",
"zh": "與『動態構圖』的語義貼近度",
"en": "closeness to a 'dynamic composition' cue"
},
"clip_max_sim": {
"token": "C",
"zh": "最貼近的構圖法則強度",
"en": "strength of the best-matching composition rule"
},
"laion_score": {
"token": "A",
"zh": "大眾審美評分",
"en": "crowd-taste aesthetic score"
},
"laion_pi_minmax": {
"token": "A",
"zh": "候選池內的美感排名",
"en": "aesthetic ranking within the candidate pool"
},
"fgaesq_score": {
"token": "A",
"zh": "專業級細粒度美學評分",
"en": "expert fine-grained aesthetic score"
},
"color_sat_mean": {
"token": "A",
"zh": "整體色彩飽和度",
"en": "overall color saturation"
},
"sam_fg_bg_sat_diff": {
"token": "A",
"zh": "主體與背景的飽和度落差",
"en": "saturation contrast between subject and background"
},
"sam_fg_bg_val_diff": {
"token": "A",
"zh": "主體與背景的明暗落差",
"en": "brightness contrast between subject and background"
},
"sam_fg_bg_lab_dist": {
"token": "A",
"zh": "主體從背景跳出的色彩對比",
"en": "how strongly the subject pops in color from the background"
},
"sam_fg_var": {
"token": "A",
"zh": "主體區域的色彩豐富度",
"en": "color richness within the subject"
},
"sam_emb_norm": {
"token": "I",
"zh": "框內的資訊量",
"en": "amount of visual information inside the crop"
},
"sam_emb_var": {
"token": "I",
"zh": "框內資訊的變化度",
"en": "variation of information inside the crop"
},
"sam_emb_kurt": {
"token": "I",
"zh": "框內資訊的集中度",
"en": "how concentrated the information is inside the crop"
},
"sam_cand_norm_mean": {
"token": "I",
"zh": "框內物件訊號的平均強度",
"en": "average object signal inside the crop"
},
"sam_cand_norm_max": {
"token": "I",
"zh": "框內最強的物件訊號",
"en": "strongest object signal inside the crop"
},
"sam_cand_norm_std": {
"token": "I",
"zh": "框內物件訊號的起伏",
"en": "spread of object signal inside the crop"
},
"sam_cand_to_outside_ratio": {
"token": "I",
"zh": "框內外資訊對比（切掉了多少）",
"en": "inside-vs-outside information contrast (how much was cut)"
},
"sam_image_total_norm": {
"token": "I",
"zh": "整張原圖的資訊量基準",
"en": "the whole image's information baseline"
},
"sam_cos_to_image": {
"token": "I",
"zh": "景別的遠近（特寫↔全景）",
"en": "shot scale from close-up to full view"
},
"sam_obj_count_in_cand": {
"token": "I",
"zh": "框內的物件數量",
"en": "number of objects inside the crop"
},
"sam_obj_top1_share": {
"token": "I",
"zh": "最大物件的佔比",
"en": "the largest object's share"
},
"sam_obj_top3_share": {
"token": "I",
"zh": "前三大物件的合計佔比",
"en": "the top-three objects' combined share"
},
"sam_obj_area_cv": {
"token": "I",
"zh": "物件之間的大小落差",
"en": "size disparity among objects"
},
"sam_dominant_ratio": {
"token": "I",
"zh": "主導物件的突出程度",
"en": "how dominant the leading object is"
},
"sam_mask_q_mean": {
"token": "I",
"zh": "物件輪廓的完整度（被切就下降）",
"en": "how cleanly objects are contained (drops when cut)"
},
"sam_mask_q_std": {
"token": "I",
"zh": "物件完整度的一致性",
"en": "consistency of object containment"
},
"sam_mask_q_margin": {
"token": "I",
"zh": "物件到框緣的安全餘裕（切殘檢查）",
"en": "object margin to the frame edge (truncation check)"
},
"sam_h_area_cv": {
"token": "I",
"zh": "前中後景的層次落差",
"en": "size contrast across depth layers"
},
"sam_h_iou_top12": {
"token": "I",
"zh": "主要場景層次之間的重疊",
"en": "overlap between the main scene layers"
},
"sam_h_cand_match_max": {
"token": "I",
"zh": "取景與場景層次結構的呼應",
"en": "how well the crop echoes the scene's layered structure"
}
}
}