anonymous347928/pcbm_metashift

--- language: - en license: mit size_categories: - 1K<n<10K task_categories: - image-classification pretty_name: Metashift subset for PCBM reproduction viewer: false dataset_info: - config_name: cherrypicked_task_1_bed_cat_dog features: - name: image dtype: image - name: label dtype: class_label: names: '0': airplane '1': bed '2': car '3': cow '4': keyboard splits: - name: train num_bytes: 28494 num_examples: 500 - name: test num_bytes: 28486 num_examples: 500 download_size: 477673284 dataset_size: 56980 - config_name: cherrypicked_task_1_bed_dog_cat features: - name: image dtype: image - name: label dtype: class_label: names: '0': airplane '1': bed '2': car '3': cow '4': keyboard splits: - name: train num_bytes: 28490 num_examples: 500 - name: test num_bytes: 28478 num_examples: 500 download_size: 477673272 dataset_size: 56968 - config_name: cherrypicked_task_2_table_books_cat features: - name: image dtype: image - name: label dtype: class_label: names: '0': beach '1': computer '2': motorcycle '3': stove '4': table splits: - name: train num_bytes: 28413 num_examples: 500 - name: test num_bytes: 28478 num_examples: 500 download_size: 477673223 dataset_size: 56891 - config_name: cherrypicked_task_2_table_books_dog features: - name: image dtype: image - name: label dtype: class_label: names: '0': beach '1': computer '2': motorcycle '3': stove '4': table splits: - name: train num_bytes: 28411 num_examples: 500 - name: test num_bytes: 28477 num_examples: 500 download_size: 477673220 dataset_size: 56888 - config_name: cherrypicked_task_2_table_cat_dog features: - name: image dtype: image - name: label dtype: class_label: names: '0': beach '1': computer '2': motorcycle '3': stove '4': table splits: - name: train num_bytes: 28477 num_examples: 500 - name: test num_bytes: 28485 num_examples: 500 download_size: 477673292 dataset_size: 56962 - config_name: cherrypicked_task_2_table_dog_cat features: - name: image dtype: image - name: label dtype: class_label: names: '0': beach '1': computer '2': motorcycle '3': stove '4': table splits: - name: train num_bytes: 28476 num_examples: 500 - name: test num_bytes: 28484 num_examples: 500 download_size: 477673290 dataset_size: 56960 - config_name: seed42_task_1_bed_cat_dog features: - name: image dtype: image - name: label dtype: class_label: names: '0': airplane '1': bed '2': car '3': cow '4': keyboard splits: - name: train num_bytes: 28498 num_examples: 500 - name: test num_bytes: 28480 num_examples: 500 download_size: 477673282 dataset_size: 56978 - config_name: seed42_task_1_bed_dog_cat features: - name: image dtype: image - name: label dtype: class_label: names: '0': airplane '1': bed '2': car '3': cow '4': keyboard splits: - name: train num_bytes: 28501 num_examples: 500 - name: test num_bytes: 28485 num_examples: 500 download_size: 477673290 dataset_size: 56986 - config_name: seed42_task_2_table_books_cat features: - name: image dtype: image - name: label dtype: class_label: names: '0': beach '1': computer '2': motorcycle '3': stove '4': table splits: - name: train num_bytes: 28434 num_examples: 500 - name: test num_bytes: 28481 num_examples: 500 download_size: 477673247 dataset_size: 56915 - config_name: seed42_task_2_table_books_dog features: - name: image dtype: image - name: label dtype: class_label: names: '0': beach '1': computer '2': motorcycle '3': stove '4': table splits: - name: train num_bytes: 28434 num_examples: 500 - name: test num_bytes: 28479 num_examples: 500 download_size: 477673245 dataset_size: 56913 - config_name: seed42_task_2_table_cat_dog features: - name: image dtype: image - name: label dtype: class_label: names: '0': beach '1': computer '2': motorcycle '3': stove '4': table splits: - name: train num_bytes: 28465 num_examples: 500 - name: test num_bytes: 28479 num_examples: 500 download_size: 477673274 dataset_size: 56944 - config_name: seed42_task_2_table_dog_cat features: - name: image dtype: image - name: label dtype: class_label: names: '0': beach '1': computer '2': motorcycle '3': stove '4': table splits: - name: train num_bytes: 28463 num_examples: 500 - name: test num_bytes: 28481 num_examples: 500 download_size: 477673274 dataset_size: 56944 --- # PCBM Metashift For the sake of reproducibility, this dataset hosts the postprocessed Metashift according to [[Yuksekgonul et al.]](https://arxiv.org/pdf/2205.15480.pdf) for the use in Post-Hoc Concept Bottleneck Models. | Config Name | Description | |---|---| | `task_1_bed_cat_dog` | Task 1: bed(cat) -> bed(dog) | | `task_1_bed_dog_cat` | Task 1: bed(dog) -> bed(cat) | | `task_2_table_books_cat` | Task 2: table(books) -> table(cat) | | `task_2_table_books_dog` | Task 2: table(books) -> table(dog) | | `task_2_table_cat_dog` | Task 2: table(cat) -> table(dog) | | `task_2_table_dog_cat` | Task 2: table(dog) -> table(cat) | The script to generate this dataset can be found at `scripts/generate.py`. You will need to download the [Metashift repo](https://github.com/Weixin-Liang/MetaShift) and the [Visual Genome dataset](https://nlp.stanford.edu/data/gqa/images.zip) as instructed in the Metashift repo.

原始信息汇总

数据集概述

基本信息

• 语言: 英语
• 许可证: MIT
• 大小类别: 1K<n<10K
• 任务类别: 图像分类
• 数据集名称: Metashift subset for PCBM reproduction

数据集配置详情

配置 cherrypicked_task_1_bed_cat_dog

• 特征:
  • image: 图像
  • label: 类别标签，包括 airplane, bed, car, cow, keyboard
• 分割:
  • train: 500个样本，28494字节
  • test: 500个样本，28486字节
• 下载大小: 477673284字节
• 数据集大小: 56980字节

配置 cherrypicked_task_1_bed_dog_cat

• 特征:
  • image: 图像
  • label: 类别标签，包括 airplane, bed, car, cow, keyboard
• 分割:
  • train: 500个样本，28490字节
  • test: 500个样本，28478字节
• 下载大小: 477673272字节
• 数据集大小: 56968字节

配置 cherrypicked_task_2_table_books_cat

• 特征:
  • image: 图像
  • label: 类别标签，包括 beach, computer, motorcycle, stove, table
• 分割:
  • train: 500个样本，28413字节
  • test: 500个样本，28478字节
• 下载大小: 477673223字节
• 数据集大小: 56891字节

配置 cherrypicked_task_2_table_books_dog

• 特征:
  • image: 图像
  • label: 类别标签，包括 beach, computer, motorcycle, stove, table
• 分割:
  • train: 500个样本，28411字节
  • test: 500个样本，28477字节
• 下载大小: 477673220字节
• 数据集大小: 56888字节

配置 cherrypicked_task_2_table_cat_dog

• 特征:
  • image: 图像
  • label: 类别标签，包括 beach, computer, motorcycle, stove, table
• 分割:
  • train: 500个样本，28477字节
  • test: 500个样本，28485字节
• 下载大小: 477673292字节
• 数据集大小: 56962字节

配置 cherrypicked_task_2_table_dog_cat

• 特征:
  • image: 图像
  • label: 类别标签，包括 beach, computer, motorcycle, stove, table
• 分割:
  • train: 500个样本，28476字节
  • test: 500个样本，28484字节
• 下载大小: 477673290字节
• 数据集大小: 56960字节

配置 seed42_task_1_bed_cat_dog

• 特征:
  • image: 图像
  • label: 类别标签，包括 airplane, bed, car, cow, keyboard
• 分割:
  • train: 500个样本，28498字节
  • test: 500个样本，28480字节
• 下载大小: 477673282字节
• 数据集大小: 56978字节

配置 seed42_task_1_bed_dog_cat

• 特征:
  • image: 图像
  • label: 类别标签，包括 airplane, bed, car, cow, keyboard
• 分割:
  • train: 500个样本，28501字节
  • test: 500个样本，28485字节
• 下载大小: 477673290字节
• 数据集大小: 56986字节

配置 seed42_task_2_table_books_cat

• 特征:
  • image: 图像
  • label: 类别标签，包括 beach, computer, motorcycle, stove, table
• 分割:
  • train: 500个样本，28434字节
  • test: 500个样本，28481字节
• 下载大小: 477673247字节
• 数据集大小: 56915字节

配置 seed42_task_2_table_books_dog

• 特征:
  • image: 图像
  • label: 类别标签，包括 beach, computer, motorcycle, stove, table
• 分割:
  • train: 500个样本，28434字节
  • test: 500个样本，28479字节
• 下载大小: 477673245字节
• 数据集大小: 56913字节

配置 seed42_task_2_table_cat_dog

• 特征:
  • image: 图像
  • label: 类别标签，包括 beach, computer, motorcycle, stove, table
• 分割:
  • train: 500个样本，28465字节
  • test: 500个样本，28479字节
• 下载大小: 477673274字节
• 数据集大小: 56944字节

配置 seed42_task_2_table_dog_cat

• 特征:
  • image: 图像
  • label: 类别标签，包括 beach, computer, motorcycle, stove, table
• 分割:
  • train: 500个样本，28463字节
  • test: 500个样本，28481字节
• 下载大小: 477673274字节
• 数据集大小: 56944字节