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 / README.md
Belall87's picture
Belall87
Upload folder using huggingface_hub
c18b08b verified
preview code
|
raw
history blame contribute delete
4.89 kB
metadata 
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 

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

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

Citation 

@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