Can you provide the .json file format for CUB_200_2011 dataset?

[dipanjyoti]

Hi,

Thank you for the excellent work. I was going through your code and encountered an issue with the dataset (.son file). I’m unable to access the Dropbox and Google Drive links for the datasets.

Could you please provide the format for the .json file, particularly for FGVC tasks, such as with the CUB_200_2011 dataset? I tried following the COCO dataset's .json file format, but it didn’t work.

Thanks!

[Haiyang-Zhou]

``import os
import json
import random

def create_split_json(data_dir, split_ratios=(0.7, 0.2, 0.1), seed=42):

# 初始化数据结构
json_data = {
    'train': {},
    'val': {},
    'test': {}
}

# 设置随机种子
random.seed(seed)

# 检查images目录
images_dir = os.path.join(data_dir, 'images')
if not os.path.exists(images_dir):
    raise ValueError(f"目录不存在: {images_dir}")

# 遍历类别文件夹
for class_name in os.listdir(images_dir):
    class_dir = os.path.join(images_dir, class_name)
    if not os.path.isdir(class_dir):
        continue  # 跳过非文件夹文件

    # 获取类别ID（这里假设文件夹名为数字开头，如 "0.cat"）
    try:
        class_id = int(class_name.split(".")[0])
    except ValueError:
        raise ValueError(f"无效的类别文件夹名: {class_name}，必须为数字开头格式（如0.cat）")

    # 获取所有图像文件
    all_images = [f for f in os.listdir(class_dir) if os.path.isfile(os.path.join(class_dir, f))]
    random.shuffle(all_images)  # 打乱顺序

    # 计算划分点
    total = len(all_images)
    train_end = int(total * split_ratios[0])
    val_end = train_end + int(total * split_ratios[1])

    # 划分数据集
    train_files = all_images[:train_end]
    val_files = all_images[train_end:val_end]
    test_files = all_images[val_end:]

    # 将路径写入对应集合
    for split, files in zip(['train', 'val', 'test'], [train_files, val_files, test_files]):
        for img_name in files:
            # 路径格式：images/类别文件夹/图片名
            img_path = os.path.join('images', class_name, img_name)
            json_data[split][img_path] = class_id

# 保存JSON文件
for split in ['train', 'val', 'test']:
    output_path = os.path.join(data_dir, f'{split}.json')
    with open(output_path, 'w') as f:
        json.dump(json_data[split], f, indent=4)

return json_data

dataset_path = r"/mnt/f/pycharm2024/pythonProject/Github/vpt/CUB_200_2011" # 替换为实际路径
create_split_json(dataset_path)

[Z-Z188]

        img_path = os.path.join('images', class_name, img_name)

改为
img_path = os.path.join('', class_name, img_name)
会更好

[LaughAwayZHY]

        img_path = os.path.join('images', class_name, img_name)

改为 img_path = os.path.join('', class_name, img_name) 会更好

Hello, have you found the corresponding hyperparameter link?