---
license: cc0-1.0
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: image
    dtype: image
  - name: label
    dtype:
      class_label:
        names:
          '0': yaleB01
          '1': yaleB02
          '2': yaleB03
          '3': yaleB04
          '4': yaleB05
          '5': yaleB06
          '6': yaleB07
          '7': yaleB08
          '8': yaleB09
          '9': yaleB10
          '10': yaleB11
          '11': yaleB12
          '12': yaleB13
          '13': yaleB15
          '14': yaleB16
          '15': yaleB17
          '16': yaleB18
          '17': yaleB19
          '18': yaleB20
          '19': yaleB21
          '20': yaleB22
          '21': yaleB23
          '22': yaleB24
          '23': yaleB25
          '24': yaleB26
          '25': yaleB27
          '26': yaleB28
          '27': yaleB29
          '28': yaleB30
          '29': yaleB31
          '30': yaleB32
          '31': yaleB33
          '32': yaleB34
          '33': yaleB35
          '34': yaleB36
          '35': yaleB37
          '36': yaleB38
          '37': yaleB39
  splits:
  - name: train
    num_bytes: 40564197
    num_examples: 2453
  download_size: 40809256
  dataset_size: 40564197
---

# **Cropped Yale Face Dataset (Grayscale)**

A clean and standardized version of the **Cropped Yale Facial Image Dataset**, containing **grayscale 168×192 cropped facial images** captured under controlled illumination conditions.
This dataset is widely used for:

* Face recognition
* Illumination-invariant modeling
* Classical computer vision research
* Autoencoders & generative models

---

## **Overview**

The **Cropped Yale Face Dataset** is derived from the original **Yale Face Database B**.

This version contains:

* **28 human subjects**
* **Frontal face images only**
* **Strong illumination variations** from many light source directions
* **Aligned, cropped, grayscale images**

Ideal for:

* Face recognition experiments
* Light normalization research
* PCA/LDA classical ML tasks
* Autoencoders, GANs, and image reconstruction tasks

---

## **Dataset Structure**

```
dataset/
│
├── yaleB01/
│   ├── yaleB01_P00A+000E+00.pgm
│   ├── yaleB01_P00A+000E+01.pgm
│   └── ...
│
├── yaleB02/
│   ├── yaleB02_P00A+000E+00.pgm
│   ├── yaleB02_P00A+000E+01.pgm
│   └── ...
│
└── ...
```

---

## **File Format**

| Property           | Value           |
| ------------------ | --------------- |
| Image size         | **168 × 192**   |
| Color mode         | **Grayscale**   |
| File type          | `.png` / `.jpg` |
| Subjects           | **28**          |
| Images per subject | ~**64**         |

---

## **Example Usage**

### **Load Images with Python (Hugging Face Datasets)**

```python
from datasets import load_dataset
import matplotlib.pyplot as plt

ds = load_dataset("YOUR_USERNAME/cropped-yale")

sample = ds["train"][0]["image"]
plt.imshow(sample, cmap="gray")
plt.axis("off")
```

---

### **TensorFlow Preprocessing Example**

```python
import tensorflow as tf

def preprocess(img):
    img = tf.image.resize(img, (192, 168))
    img = tf.cast(img, tf.float32) / 255.0
    return img
```

---

### **PyTorch Example**

```python
from torchvision import transforms

transform = transforms.Compose([
    transforms.Resize((192, 168)),
    transforms.ToTensor()
])
```

---

## **Applications**

### **Face Recognition**

Train classical or modern models:

* Eigenfaces
* Fisherfaces
* SVM classifiers
* CNN-based architectures

### **Illumination-Invariant Face Analysis**

Evaluate model robustness under extreme lighting shifts.

### **Dimensionality Reduction**

Perfect for:

* PCA
* LDA
* Linear subspace modeling

### **Autoencoders / GANs**

Great for:

* Reconstruction
* Denoising
* Generative modeling

---

## **Sample Images**

![Example](https://files.catbox.moe/fjxmf6.jpg)

---

## **Download / Use**

If you're viewing this on **Hugging Face**, simply click:

> **Use dataset → Load in Python**

Or install via:

```python
load_dataset("YOUR_USERNAME/cropped-yale")
```

---

## **License & Citation**

This dataset is derived from:

**Yale Face Database B** and the **Cropped Yale Dataset**, prepared by Yale University researchers.

If you use this dataset in academic work, please cite:

> Georghiades, A. S., Belhumeur, P. N., & Kriegman, D. J.
> *From Few Pixels to the Illumination Cone Model*.
> IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2001.

All images are provided **for research and academic purposes only**.

---

## **Acknowledgements**

Special thanks to **Yale University** for releasing the original dataset and supporting reproducible computer vision research.

---