Hugging Face's logo

Belall87
/
Cat-Emotion-Classification-with-CNN

Image Classification
Keras
English
vgg16-transfer-learning
cat-emotion
vgg16
transfer-learning
cnn
tensorflow
computer-vision
depi
Model card Files
xet
Community
Cat-Emotion-Classification-with-CNN
File size: 4,893 Bytes 
c18b08b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
 
---

language: en
license: mit
library_name: keras
pipeline_tag: image-classification
tags:
  - cat-emotion
  - vgg16
  - transfer-learning
  - cnn
  - tensorflow
  - keras
  - image-classification
  - computer-vision
  - depi
datasets:
  - custom
metrics:
  - accuracy
---


# 🐱 Cat Emotion Classification with VGG16

A deep learning model for classifying cat emotions from images using **VGG16 transfer learning**, achieving **82.19% validation accuracy** across 5 emotion classes.

## Model Description

This model classifies cat facial expressions into one of five emotional categories. It uses VGG16 (pre-trained on ImageNet) as a feature extractor with a custom classification head featuring GlobalAveragePooling2D, BatchNormalization, and two Dense layers.

**Part of:** Digital Egypt Pioneers Initiative (DEPI) Machine Learning Internship

| Property | Value |
|----------|-------|
| **Architecture** | VGG16 + GAP + BN + Dense(512, 256) |
| **Input Size** | 224×224×3 RGB |
| **Output** | 5 classes (softmax) |
| **Parameters** | ~15.3M (trainable: ~2.1M) |
| **Framework** | TensorFlow / Keras |
| **Accuracy** | 82.19% |

## Classes

| Index | Class | Description |
|-------|-------|-------------|
| 0 | `angry` | Cat displaying angry expressions |
| 1 | `normal` | Cat in a neutral/normal state |
| 2 | `rested` | Cat in a relaxed/resting state |
| 3 | `sad` | Cat displaying sad expressions |
| 4 | `surprised` | Cat displaying surprised expressions |

## Quick Start

```python

from tensorflow.keras.models import load_model

import numpy as np

from PIL import Image



# Load model

model = load_model('best_vgg16_model.keras')



# Predict

img = Image.open('cat.jpg').resize((224, 224))

img_array = np.expand_dims(np.array(img) / 255.0, axis=0)



prediction = model.predict(img_array)

classes = ['angry', 'normal', 'rested', 'sad', 'surprised']



predicted_class = classes[np.argmax(prediction)]

confidence = np.max(prediction)

print(f"Emotion: {predicted_class} ({confidence:.2%})")

```

## Training Details

### Architecture (V2 — Final)

```

VGG16 (ImageNet, last 8 layers unfrozen)

    ↓

GlobalAveragePooling2D

    ↓

BatchNormalization

    ↓

Dense(512, ReLU) → Dropout(0.5)

    ↓

BatchNormalization

    ↓

Dense(256, ReLU) → Dropout(0.3)

    ↓

Dense(5, Softmax)

```

### Training Strategy

| Phase | Epochs | Learning Rate | Description |
|-------|--------|--------------|-------------|
| Phase 1 — Feature Extraction | 15 | 1e-4 | All VGG16 layers frozen |
| Phase 2 — Fine-Tuning | 25 | 1e-5 | Last 8 VGG16 layers unfrozen |

- **Optimizer:** Adam
- **Loss:** Categorical Crossentropy
- **Class Weights:** Balanced (computed with sklearn)
- **Callbacks:** EarlyStopping (patience=5), ReduceLROnPlateau, ModelCheckpoint

### Data Augmentation
- Rotation: 30°
- Width/Height shift: 0.25
- Zoom: 0.25
- Horizontal flip
- Brightness: [0.8, 1.2]
- Shear: 0.15

## Performance

### Model Comparison

| Version | Input | Architecture | Accuracy |
|---------|-------|-------------|----------|
| V1 (Baseline) | 128×128 | Flatten → Dense(256) | 81.18% |
| **V2 (Final)** | **224×224** | **GAP → BN → Dense(512,256)** | **82.19%** |

### Dataset

- **Training:** ~11,513 images
- **Validation:** ~2,880 images
- **Source:** [Kaggle — Cats Data Set](https://www.kaggle.com/datasets/bilalmahmoud/cats-data-set)

## Files

- `best_vgg16_model.keras` — Trained VGG16 model (V1, 134MB)
- `README.md` — This model card

## Intended Use

- **Primary use:** Classifying cat emotions from images
- **Users:** Researchers, students, pet tech developers
- **Out of scope:** Other animal species, real-time production systems

## Limitations

- Trained on a specific cat emotion dataset — may not generalize to all cat breeds or lighting conditions
- 5 emotion categories only — does not cover all possible feline emotional states
- Best results with clear, well-lit frontal images of cats

## Links

- 🐱 **GitHub:** [Bolaal/Cat-Emotion-Classification-with-CNN](https://github.com/Bolaal/Cat-Emotion-Classification-with-CNN)
- 📊 **Dataset:** [Kaggle — Cats Data Set](https://www.kaggle.com/datasets/bilalmahmoud/cats-data-set)

## Citation

```bibtex

@misc{cat-emotion-vgg16-2025,

  author = {Belal Mahmoud Hussien},

  title = {Cat Emotion Classification with VGG16},

  year = {2025},

  publisher = {Hugging Face},

  url = {https://huggingface.co/Belall87/Cat-Emotion-Classification-with-CNN}

}

```

## Author

**Belal Mahmoud Hussien**
- 📧 Email: belalmahmoud8787@gmail.com
- 💼 LinkedIn: [belal-mahmoud-husien](https://linkedin.com/in/belal-mahmoud-husien)
- 🐱 GitHub: [@Bolaal](https://github.com/Bolaal)
- 🤗 Hugging Face: [@Belall87](https://huggingface.co/Belall87)