oxford-iiit-pet-clean
收藏资源简介：
Oxford-IIIT-Pet Cleaned 数据集是原始 Oxford-IIIT Pet 数据集的清理和增强版本，专为细粒度图像分类任务设计。该版本解决了原始数据集中的常见质量问题，如重复文件、损坏文件以及训练/测试集之间的泄漏问题，同时保留了原始数据结构。数据集包含37个宠物类别，每个类别约有200张图像，图像在尺度、姿态和光照方面存在较大变化。 数据集包含四个分割：train_clean（80%的训练集，移除所有关键问题）、val_clean（20%的训练集，同样移除关键问题）、test_clean（移除内部重复和损坏文件）和test_original（原始测试集，未更改行但增加了问题标注）。 数据字段包括：image（图像）、label（分类标签，0-36）、image_id（唯一标识符）、label_cat_dog（猫或狗的分类标签，0或1）以及issues（标注潜在问题的列表）。issues列可能包含文件损坏、健康问题、质量问题、标签错误、内部重复或泄漏等标注。 数据集适用于细粒度图像分类任务和数据质量研究，采用CC-BY-SA 4.0许可。
The Oxford-IIIT-Pet Cleaned dataset is a cleaned and enhanced version of the original Oxford-IIIT Pet Dataset, specifically designed for fine-grained image classification tasks. This version addresses common quality issues in the original dataset, including duplicate files, corrupted files, and data leakage between the training and test splits, while preserving the original data structure. The dataset contains 37 pet categories, with approximately 200 images per category, and the images exhibit significant variations in scale, pose, and lighting conditions. The dataset includes four splits: train_clean (80% of the training set with all critical issues removed), val_clean (20% of the training set with critical issues also removed), test_clean (internal duplicates and corrupted files removed), and test_original (the original test split, unchanged but with added issue annotations). The dataset comprises the following fields: image (the image file), label (classification label ranging from 0 to 36), image_id (unique identifier), label_cat_dog (binary classification label for cat or dog, 0 or 1), and issues (a list of annotated potential issues). The issues column may contain annotations such as file corruption, health-related issues, quality defects, label errors, internal duplicates, or data leakage. This dataset is suitable for fine-grained image classification tasks and data quality research, and is released under the CC-BY-SA 4.0 license.
