Datasets:
File size: 3,052 Bytes
e7882b2 daa6a83 e7882b2 eae59b9 e7882b2 44704c2 e7882b2 eae59b9 e7882b2 eae59b9 e7882b2 eae59b9 e7882b2 eae59b9 e7882b2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 |
---
license: cc-by-4.0
task_categories:
- image-classification
language:
- en
tags:
- benchmark
- image-classification
- out-of-distribution
- robustness
- sensor-control
- light-control
- real-photo
size_categories:
- 100K<n<1M
---
# 📸 ImageNet-ES
Unlike conventional robustness benchmarks that rely on digital perturbations, we directly capture **202k images** by using a real camera in a controllable testbed. **The dataset presents a wide range of covariate shifts caused by variations in light and camera sensor factors.**
[📄 Read the paper (CVPR 2024)](https://openaccess.thecvf.com/content/CVPR2024/html/Baek_Unexplored_Faces_of_Robustness_and_Out-of-Distribution_Covariate_Shifts_in_Environment_CVPR_2024_paper.html)
<img align="center" src="https://raw.githubusercontent.com/Edw2n/ImageNet-ES/main/supples/ImageNet-ES.jpg" width="800">
---
### 🗂️ ImageNet-ES Strucuture
```
ImageNet-ES
├── es-train
│ └── tin_no_resize_sample_removed
│ # 8K original validation samples of Tiny-ImageNet without references
├── es-val
│ ├── auto_exposure # 10K = 1K reference samples * 2 environments * 5 shots
│ ├── param_control # 128K = 1K reference samples * 2 environments * 64 shots
│ └── sampled_tin_no_resize # reference samples (1K)
├── es-test
├── auto_exposure # 10K = 1K reference samples * 2 environments * 5 shots
├── param_control # 54K = 1K reference samples * 2 environments * 27 shots
└── sampled_tin_no_resize2 # reference samples (1K)
```
The main paper and the appendix detail the dataset specifications and present analyses on covariate shifts, robustness evaluations, and qualitative insights.
---
### 🎛️ ES-Studio
To compensate the missing perturbations in current robustness benchmarks, we construct a new testbed, **ES-Studio** (**E**nvironment and camera **S**ensor perturbation **Studio**). It can control physical light and camera sensor parameters during data collection.
<img align="center" src="https://raw.githubusercontent.com/Edw2n/ImageNet-ES/main/supples/Testbed.png" width="800">
<img align="center" src="https://raw.githubusercontent.com/Edw2n/ImageNet-ES/main/supples/Testbed_actual.jpg" width="800">
---
### 🖥️ Download from terminal
To download the dataset directly from your terminal using **`wget`:**
```bash
wget https://huggingface.co/datasets/Edw2n/ImageNet-ES/resolve/main/ImageNet-ES.zip
```
---
### 🔍 More Exploration
Visit our paper repository: [🔗 ImageNet-ES GitHub Repository](https://github.com/Edw2n/ImageNet-ES)
---
### 📜 Citation
```bibtex
@InProceedings{Baek_2024_CVPR,
author = {Baek, Eunsu and Park, Keondo and Kim, Jiyoon and Kim, Hyung-Sin},
title = {Unexplored Faces of Robustness and Out-of-Distribution: Covariate Shifts in Environment and Sensor Domains},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2024},
pages = {22294--22303}
}
``` |