Datasets:
File size: 2,919 Bytes
0e3d03f c3ffd72 0e3d03f 1e26ec2 0e3d03f |
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 71 72 73 |
---
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 Diverse
**ImageNet-ES Diverse** is a benchmark dataset of **192,000 real-world images** captured with a physical camera in a controlled testbed (ES-Studio Diverse), under various sensor parameters and lighting conditions. It complements the original [ImageNet-ES](https://huggingface.co/datasets/Edw2n/ImageNet-ES) by introducing more diverse and realistic covariate shifts for robustness evaluation.
This dataset is introduced in the ICLR 2025 paper: *Adaptive Camera Sensor for Vision Models*.
[📄 Read the paper (ICLR 2025)](https://openreview.net/pdf?id=He2FGdmsas)
<img align="center" src="https://raw.githubusercontent.com/Edw2n/Lens/main/ImageNet-ES-Diverse-example.png" width="800">
---
### 🗂️ ImageNet-ES Diverse Strucuture
```
ImageNet-ES-Diverse
├── es-train (8K)
│ └── tin_no_resize_sample_removed
│ # 8K original validation samples of Tiny-ImageNet without references in ImageNet-ES
├── es-diverse-test (193K)
├── auto_exposure (30K) = 1K * 6 environments *5 shots
├── param_control (162K) = 1K * 6 environments * 27 shots
└── sampled_tin_no_resize2 # reference samples (1K)
```
The main paper and the appendix detail the dataset specifications and present analyses on adaptive sensing and qualitative insights.
---
### 🎛️ ES-Studio Diverse
To address the limitations of existing robustness benchmarks, including *ImageNet-ES*, we customized *ES-Studio* into *ES-Studio Diverse* by replacing the reference display with printed images captured from a screen. While maintaining control over lighting and sensor parameters as in *ES-Studio*, this customized setup and broader lighting variations enable the collection of data that is complementary to *ImageNet-ES*.
<img align="center" src="https://raw.githubusercontent.com/Edw2n/Lens/main/ES-Studio-Diverse.png" width="400">
<img align="center" src="https://raw.githubusercontent.com/Edw2n/Lens/main/testbed-diverse-actual-.png" width="400">
---
### 🖥️ Download from terminal
To download the dataset directly from your terminal using **`wget`:**
```bash
wget https://huggingface.co/datasets/Edw2n/ImageNet-ES-Diverse/resolve/main/ImageNet-ES-Diverse.zip
```
---
### 🔍 More Exploration
Visit our paper repository: [🔗 Lens GitHub Repository](https://github.com/Edw2n/Lens)
---
### 📜 Citation
```bibtex
@inproceedings{
baek2025adaptive,
title={Adaptive Camera Sensor for Vision Models},
author={Eunsu Baek and Sung-hwan Han and Taesik Gong and Hyung-Sin Kim},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=He2FGdmsas}
}
``` |