facebook/PUG_ImageNet

---
license: cc-by-nc-4.0
dataset_info:
features:
- name: image
dtype: image
- name: world_name
dtype: string
- name: character_name
dtype: string
- name: character_label
dtype: string
- name: character_rotation_yaw
dtype: int64
- name: character_rotation_roll
dtype: int64
- name: character_rotation_pitch
dtype: int64
- name: character_scale
dtype: float64
- name: camera_roll
dtype: int64
- name: camera_pitch
dtype: int64
- name: camera_yaw
dtype: int64
- name: character_texture
dtype: string
- name: scene_light
dtype: string
splits:
- name: train
num_bytes: 29382707151.112
num_examples: 88328
download_size: 29358745565
dataset_size: 29382707151.112
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

## PUG: ImageNet
The PUG: ImageNet dataset contains 88,328 pre-rendered images based on Unreal Engine using 724 assets representing 151 ImageNet classes with 64 environments, 7 sizes, 9 textures, 18 different camera orientations, 18 different character orientations and 7 light intensities. In contrast to PUG: Animals, PUG: ImageNet was created by varying only a single factor at a time (which explains the lower number of images than PUG: Animals despite using more factors). The main purpose of this dataset is to provide a novel, useful benchmark, paralleling ImageNet, but for fine-grained evaluation of the robustness of image classifiers, along several factors of variation.

## LICENSE
The datasets are distributed under the CC-BY-NC, with the addenda that they should not be used to train Generative AI models.

## Citing PUG
If you use one of the PUG datasets, please cite:
```
@misc{bordes2023pug,
title={PUG: Photorealistic and Semantically Controllable Synthetic Data for Representation Learning},
author={Florian Bordes and Shashank Shekhar and Mark Ibrahim and Diane Bouchacourt and Pascal Vincent and Ari S. Morcos},
year={2023},
eprint={2308.03977},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```

## To learn more about the PUG datasets:
Please visit the [website](https://pug.metademolab.com/) and the [github](https://github.com/facebookresearch/PUG)

原始信息汇总

PUG: ImageNet 数据集概述

数据集信息

• 特征列表：

  • image: 图像数据
  • world_name: 字符串，世界名称
  • character_name: 字符串，角色名称
  • character_label: 字符串，角色标签
  • character_rotation_yaw: 整数，角色偏航旋转
  • character_rotation_roll: 整数，角色滚动旋转
  • character_rotation_pitch: 整数，角色俯仰旋转
  • character_scale: 浮点数，角色比例
  • camera_roll: 整数，相机滚动
  • camera_pitch: 整数，相机俯仰
  • camera_yaw: 整数，相机偏航
  • character_texture: 字符串，角色纹理
  • scene_light: 字符串，场景光照

• 数据分割：

  • train: 训练集，包含 88328 个样本，总大小为 29382707151.112 字节

• 数据集大小：

  • 下载大小：29358745565 字节
  • 实际大小：29382707151.112 字节

• 配置：

  • default 配置，包含训练集文件路径 data/train-*

数据集描述

• 内容：PUG: ImageNet 数据集包含 88,328 张预渲染图像，基于 Unreal Engine 使用 724 个资产，代表 151 个 ImageNet 类别，涉及 64 个环境、7 种尺寸、9 种纹理、18 种不同相机方向、18 种不同角色方向和 7 种光照强度。
• 目的：该数据集的主要目的是提供一个新颖、有用的基准，类似于 ImageNet，但用于细粒度评估图像分类器的鲁棒性，沿着多个变量因素。

许可证

• 数据集在 CC-BY-NC 许可证下分发，附加条件是不得用于训练生成式 AI 模型。