Datasets:
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
task_categories:
|
| 4 |
+
- image-classification
|
| 5 |
+
tags:
|
| 6 |
+
- domain-generalization
|
| 7 |
+
- computer-vision
|
| 8 |
+
- benchmark
|
| 9 |
+
pretty_name: PACS
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# PACS Dataset
|
| 13 |
+
|
| 14 |
+
## Overview
|
| 15 |
+
|
| 16 |
+
PACS is a benchmark dataset for **domain generalization** in image classification,
|
| 17 |
+
introduced in "Deeper, Broader and Artier Domain Generalization" (Li et al., ICCV 2017).
|
| 18 |
+
|
| 19 |
+
It contains **9,991 images** across **4 domains** and **7 object categories**,
|
| 20 |
+
with significantly larger domain shift than prior benchmarks like VLCS —
|
| 21 |
+
averaging a 20.2% cross-domain performance drop versus 10.0% for VLCS.
|
| 22 |
+
|
| 23 |
+
## Domains
|
| 24 |
+
|
| 25 |
+
| Domain | Description |
|
| 26 |
+
|---|---|
|
| 27 |
+
| **P** — Photo | Real photographs |
|
| 28 |
+
| **A** — Art Painting | Artistic paintings |
|
| 29 |
+
| **C** — Cartoon | Cartoon-style illustrations |
|
| 30 |
+
| **S** — Sketch | Hand-drawn sketches |
|
| 31 |
+
|
| 32 |
+
## Classes
|
| 33 |
+
|
| 34 |
+
7 categories: **dog, elephant, giraffe, guitar, horse, house, person**
|
| 35 |
+
|
| 36 |
+
## Dataset Statistics
|
| 37 |
+
|
| 38 |
+
| Domain | Images |
|
| 39 |
+
|---|---|
|
| 40 |
+
| Photo | ~1,670 |
|
| 41 |
+
| Art Painting | ~2,048 |
|
| 42 |
+
| Cartoon | ~2,344 |
|
| 43 |
+
| Sketch | ~3,929 |
|
| 44 |
+
| **Total** | **9,991** |
|
| 45 |
+
|
| 46 |
+
## Usage
|
| 47 |
+
|
| 48 |
+
The standard evaluation protocol is **leave-one-domain-out**: train on 3 domains,
|
| 49 |
+
test on the held-out domain. This yields 4 cross-domain tasks:
|
| 50 |
+
- Train on A, C, S → Test on P
|
| 51 |
+
- Train on P, C, S → Test on A
|
| 52 |
+
- Train on P, A, S → Test on C
|
| 53 |
+
- Train on P, A, C → Test on S
|
| 54 |
+
|
| 55 |
+
## Citation
|
| 56 |
+
|
| 57 |
+
```bibtex
|
| 58 |
+
@inproceedings{li2017deeper,
|
| 59 |
+
title={Deeper, Broader and Artier Domain Generalization},
|
| 60 |
+
author={Li, Da and Yang, Yongxin and Song, Yi-Zhe and Hospedales, Timothy M},
|
| 61 |
+
booktitle={ICCV},
|
| 62 |
+
year={2017}
|
| 63 |
+
}
|
| 64 |
+
```
|
| 65 |
+
|
| 66 |
+
## Uploaded By
|
| 67 |
+
|
| 68 |
+
Mohammed Azeez Khan — used for domain generalization experiments at
|
| 69 |
+
Carnegie Mellon University (EEG P300, motor imagery, fMRI neuroimaging).
|