🧠 Image Classification with CNN (CIFAR-10)

👤 Maxim Merkurev

🏫 Far Eastern Federal University, 2025

---

📌 About the project

This project focuses on designing and training a Convolutional Neural Network (CNN) to classify images from the CIFAR-10 dataset. The main emphasis is on exploring CNN architecture and the impact of batch size and number of epochs on model accuracy.

🔗 Colab Version:

Local launch via JupyterLab is also supported.

---

🖼 CIFAR-10: Dataset Description

CIFAR-10 contains:

• 60 000 color images 32×32
• 10 classes:
  airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck
• Training set: 50 000 images
• Test set: 10 000 images

CIFAR-10 is widely used in research and competitions as a benchmark for testing computer vision models.


🧠 What is a CNN?

Convolutional Neural Networks (CNNs) are a type of architecture designed specifically for image processing:

• 🎯 Detect local patterns (edges, shapes)
• 🧱 Reduce parameter count compared to MLPs
• 🔁 Reuse filters across the image
• ⬇ Robust to translation and scaling

---

🧮 Model Architecture

Block	Components
Input	32×32×3 (RGB image)
Conv 1	64 filters (5×5) → BatchNorm → ReLU → MaxPool (2×2)
Conv 2	128 filters (3×3) → BatchNorm → ReLU
Conv 3	256 filters (3×3) → BatchNorm → ReLU → MaxPool (2×2)
Flatten	Transition to fully connected layers
FC 1	1024 neurons → ReLU → Dropout(0.5)
FC 2	1024 neurons → ReLU → Dropout(0.5)
FC 3	512 neurons → ReLU
Output	10 neurons → Softmax (for classification)

---

🚀 Installation (for local use)

1. Clone the repository:

git clone https://github.com/Bit-Maximum/CNN-for-CIFAR.git
cd CNN-for-CIFAR

2. Install dependencies:

pip install -r requirements.txt

3. Run the project:

jupyter lab run.ipynb

---

📈 Results

• 📊 Test accuracy: 78.6%
• 📉 Training graphs (available in Colab/report):
  • Smooth decrease in loss
  • Steady increase in accuracy
• 📌 Conclusion: The model shows confident learning and can be further improved by increasing the number of epochs.

🔍 Some confusion observed between similar classes (e.g., cat vs dog).

Example predictions from the trained model:

📚 Theoretical Background

Covered topics include:

• 📦 CIFAR-10 as a benchmark image classification dataset
• 🧠 CNN layers and pooling operations
• 🧮 BatchNorm, Dropout, ReLU/Softmax activations
• 🔁 Effect of training parameters like batch size and number of epochs