speed commited on
Commit
93e5b33
·
verified ·
1 Parent(s): 00245e0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +51 -17
README.md CHANGED
@@ -25,26 +25,60 @@ configs:
25
  ---
26
 
27
  # WAON-Bench: Japanese Cultural Image Classification Dataset
28
- WAON-Bench is a small-scale, manually curated image classification dataset designed to benchmark models on visual understanding of Japanese culture.
29
- It consists of 270 images, each representing a distinct category related to Japanese life, landscape, architecture, or traditions.
30
 
31
 
32
- ## Statistics
33
- - Total Images: 270
34
- - Total Categories: 270 (1 image per category)
35
- - Language: Japanese (category names)
36
-
37
- ## Category Examples
38
- Each category corresponds to a single image, chosen to clearly and uniquely illustrate the concept. Examples include:
39
- - 茶畑 (Tea plantation)
40
- - 雪国の街並み (Snow country townscape)
41
- - 漁港 (Fishing port)
 
 
 
 
 
 
 
 
 
42
  ```
43
- ['茶畑', '雪国の街並み', '漁港', '砂防ダム', '灯籠', '石垣', '田園風景', '屋上庭園', '自動販売機', '駅前ロータリー', '商店街', '能舞台', '狛犬', '鳥居', '石庭', '茅葺き屋根', '書院造', '茶室', '提灯', '神輿', '御朱印', '花見', '節分', '餅つき', '屋台', '盆踊り', '紅葉', '花火大会', '桜吹雪', '風鈴', '雪吊り', '新幹線', '路面電車', '軽トラ', 'カブ(原付バイク)', '観光人力車', '特急列車の車内販売', '地下鉄改札口', 'こけし', 'プリクラ機', '折り紙', '鯉のぼり', '習字道具', 'カラオケ', '招き猫', '五稜郭', '小樽運河', 'ねぶた', '松島', '東京ディズニーランド', 'ユニバーサルスタジオジャパン', '原爆ドーム', '富士山', '黒部ダム', '兼六園', '天橋立', '皇居', '大阪城', '名古屋城', '姫路城', '東京スカイツリー', '東京タワー', '東大寺', '浅草寺', '平等院鳳凰堂', '伏見稲荷大社', '清水寺', '金閣寺', '鎌倉大仏', '桜島', '首里城', '縄文土器', '弥生土器', '埴輪', '銅鐸', '前方後円墳', '浮世絵', 'そろばん', '花魁', '妖怪', '風呂敷', '足袋', 'ランドセル', '扇子', 'ガソリンスタンド', 'コンビニ', 'スーパーマーケット', 'バス停', 'マンション', '交番', '公園', '図書館', '学校', '映画館', '消防署', '空港', '郵便局', '駐車場', 'つばき', 'アサガオ', 'アジサイ', 'キク', 'キンモクセイ', 'タンポポ', 'チューリップ', '菜の花', 'ハイビスカス', 'バラ', 'ヒマワリ', 'マリーゴールド', 'ラベンダー', '彼岸花', '杉', 'マグロ', 'いくら', 'いちご', 'いなり寿司', 'うどん', 'おでん', 'おにぎり', 'お好み焼き', 'たこ焼き', 'きんぴらごぼう', 'しゃぶしゃぶ', 'すき焼き', 'そば', 'ちゃんぽん', 'ちらし寿司', 'ひつまぶし', '味噌汁', '天ぷら', 'ガンダム', 'ポケモン', 'ドラえもん', 'アンパンマン', '歌舞伎', '道の駅', '棚田', '防波堤', '橋梁', '首都高のジャンクション', '鉄塔(送電鉄塔)', '町家(京都の町並み)', '七五三', '節句人形(雛人形・五月人形)', '破魔矢', '熊手', '書き初め', '獅子舞', '鯛車', '郵便ポスト(丸型・赤いタイプ)', '商売繁盛の絵馬', 'パチンコ屋', 'ゲーセンのクレーンゲーム', '地蔵盆', '雪まつり', 'サンマの水揚げ', 'ほたる(ホタル観賞)', '秋葉原の電気街', 'メイドカフェ', 'プラモデル専門店', 'ご当地マンホール', '痛車', '地元ゆるキャラ', '修行僧(比叡山など)', 'お遍路さん(白装束+菅笠)', '鐘つき(除夜の鐘)', 'ビルの屋上給水タンク', '高速道路の高架下', 'ロープウェイ', '千羽鶴', 'だるま落とし', '藁のしめ縄', '洗濯物(ベランダに干された布団)', '敷布団(畳の上に敷いた状態)', 'ちゃぶ台', 'ガチャガチャ', 'カプセルホテルの内部', '電車の吊り広告', '牛丼チェーン店のカウンター席', 'スーパーの試食コーナー', '相撲大会', 'きつねの嫁入り行���', 'よさこい祭り', '蜂の巣駆除', 'ワカメ干し', '山火事防止の看板', '精進料理', '線香と数珠', '地蔵に赤い前掛け', 'うなぎの蒲焼', '三味線', '結婚式', '知床', '勾玉', '零戦', '戦艦大和', '原爆', '火炎放射器', '松茸', '椎茸', 'たけのこ', '鳥取砂丘', '山内丸山遺跡', '鴨川', '幕張メッセ', '京都タワー', 'トミカ', 'こたつ', 'シーサー', '厳島神社', 'かるた', '将棋', '茄子', 'キジ', '鷲', '蚊取り線香', 'お年玉', 'のれん', 'うちわ', 'しらす', '競馬', '競輪', '競艇', '花札', '竹とんぼ', 'おはじき', '竹馬', '琵琶', '尺八', '囲炉裏', '納屋', '風呂桶', '囲碁盤', '日本刀', '木造の橋(太鼓橋)', '不動明王像', '鐘楼門', '御神木', '灯籠流し', 'ししおどし', '牛車', 'けん玉', '市民プール', '瓶入りラムネ', '虫かごとクワガタ', '雨上がりのカタツムリ', '防災倉庫', '津波', '噴火', '石垣島のサンゴ礁', 'ゴーヤー畑', '赤べこ', '通天閣', '流氷観光船', 'サロベツ湿原', '別府の地獄めぐり', 'ランタンフェスティバル', '高床倉庫', '黒電話', 'かき氷', '金魚すくい', '射的', 'りんご飴', 'わたあめ', 'チョコバナナ', 'ベビーカステラ', '蓄音機', '枕草子', '法隆寺']
44
  ```
45
 
46
- ## Data Collection Process
47
- Categories were manually selected to cover a wide range of cultural, natural, and urban themes in Japan.
48
- For each category, a single representative image was retrieved using Google Image Search.
49
- Selection criteria emphasized clarity, cultural specificity, and low ambiguity (i.e., images that depict only one clear concept).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
50
 
 
25
  ---
26
 
27
  # WAON-Bench: Japanese Cultural Image Classification Dataset
28
+ WAON-Bench is a small-scale, manually curated image classification dataset designed to benchmark Vision-Language models on the visual understanding of Japanese culture.
29
+ It consists of 385 images, each corresponding to a distinct class, capturing a wide range of cultural, natural, and everyday elements of Japanese life.
30
 
31
 
32
+ ## Data Collection Pipeline
33
+ We followed the pipeline below to construct the dataset:
34
+
35
+ 1. **Class Definition**: A total of 385 class names were manually defined and grouped into eight top-level categories:
36
+ animal, building, event, everyday, food, nature, scenery, and tradition.
37
+ 2. **Image Selection**: For each class, a representative image was manually retrieved using Google Image Search. \
38
+ Images were selected based on the following criteria:
39
+ - The image should clearly represent the intended class.
40
+ - It should not contain elements that could be easily confused with other classes.
41
+
42
+ ## Dataset Format
43
+ Each sample includes:
44
+
45
+ - `class`: Class name
46
+ - `url`: Image URL
47
+ - `category`: Class category
48
+ - `jpg`: PIL Image object
49
+
50
+ Example:
51
  ```
52
+ {'class': '犬', 'url': 'https://www.woodtec.co.jp/products/lineup/flooring/fordog/wp/wp-content/uploads/2024/12/67-1-1024x683.jpg', 'category': 'animal', 'jpg': <PIL.PngImagePlugin.PngImageFile image mode=RGB size=1024x683 at 0x7FC03F469700>}
53
  ```
54
 
55
+ ## Statistics
56
+
57
+ - **Class num per category**
58
+ | category | class |
59
+ |:-----------|--------:|
60
+ | animal | 41 |
61
+ | building | 39 |
62
+ | event | 30 |
63
+ | everyday | 34 |
64
+ | food | 55 |
65
+ | nature | 26 |
66
+ | scenery | 85 |
67
+ | tradition | 75 |
68
+ | total | 385 |
69
+
70
+ - **Example Class Names per Category**
71
+ |category | class names|
72
+ |:-----------|--------:|
73
+ | animal | '柴犬', 'エゾシカ', 'ニホンカモシカ', 'イノシシ', 'タヌキ', ...|
74
+ | building | '鳥居', '茶室', '合掌造り', '町家', '和室', '縁側', ...|
75
+ | event | '花見', '花火大会', '盆踊り', '運動会', '卒業式', '成人式', ...|
76
+ | everyday | 'カラオケ', '温泉', '屋台', '洗濯物', 'ランドセル', ...|
77
+ | food | '茄子', 'しらす', 'ラーメン', '焼き鳥', '焼肉', '白米', '弁当', 'カレーライス', ...|
78
+ | nature | '桜', '梅', '藤', '牡丹', 'つばき', 'アサガオ', 'アジサイ', '噴火', ...|
79
+ | scenery | '茶畑', '雪国の街並み', '漁港', '砂防ダム', '石垣', '自動販売機', ...|
80
+ | tradition| '華道', '書道', '剣道', '柔道', '弓道', ...|
81
+
82
+
83
+
84