aijadugar/cifar-10-vgg16

VGG16 on CIFAR-10 using PyTorch

This project implements the VGG16 architecture from scratch using PyTorch and trains it on the CIFAR-10 image classification dataset.

Features

• VGG16 architecture implemented from scratch
• CIFAR-10 training pipeline
• Data normalization
• SGD optimizer with Momentum
• Learning Rate Scheduler
• Model evaluation
• Inference on test images
• Save and load trained model

---

Dataset

CIFAR-10 consists of 60,000 RGB images of size 32×32 belonging to 10 classes.

Classes:

• Airplane
• Automobile
• Bird
• Cat
• Deer
• Dog
• Frog
• Horse
• Ship
• Truck

Dataset path:

/kaggle/input/datasets/bariankitvinod/cifar-10/data

---

Model Architecture

Input
│
├── Conv(64)
├── Conv(64)
├── MaxPool
│
├── Conv(128)
├── Conv(128)
├── MaxPool
│
├── Conv(256)
├── Conv(256)
├── Conv(256)
├── MaxPool
│
├── Conv(512)
├── Conv(512)
├── Conv(512)
├── MaxPool
│
├── Conv(512)
├── Conv(512)
├── Conv(512)
├── MaxPool
│
├── Flatten
├── Linear(512)
├── ReLU
├── Dropout
├── Linear(512)
├── ReLU
├── Dropout
└── Linear(10)

---

Training Configuration

Loss Function

• CrossEntropyLoss

Optimizer

• SGD
• Learning Rate = 0.01
• Momentum = 0.9
• Weight Decay = 5e-4

Learning Rate Scheduler

• StepLR
• Step Size = 10
• Gamma = 0.1

Epochs

• 50

---

Expected Performance

Typical Test Accuracy

Test Accuracy : 10.00%

---

Run

Execute all notebook cells sequentially.

---

Project Structure

VGG16/
│
├── vgg16.ipynb
├── README.md
├── requirements.txt
├── config.py
└── vgg16_cifar10.pth

---

Technologies Used

• Python
• PyTorch
• Torchvision
• NumPy
• Matplotlib
• tqdm