Image Classification using CNN (CIFAR-10)

This project implements a Convolutional Neural Network (CNN) using PyTorch to classify images into 10 different categories.
The model is trained and evaluated on the standard CIFAR-10 dataset.

---

Dataset

The model uses the CIFAR-10 dataset, which contains:

• 60,000 color images
• Image size: 32 × 32 pixels
• 10 classes (6,000 images per class)

Classes:

['plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']

---

Project Structure

├── image_classfication.ipynb   # Main notebook (training + evaluation + inference)
├── first_classification_model.pth  # Saved trained model
├── 11.png, 22.png, 33.png, image.png  # Sample test images

---

Requirements

Make sure you have the following installed:

• Python 3.x
• PyTorch
• Torchvision
• NumPy
• Pillow (PIL)

Install dependencies:

pip install torch torchvision numpy pillow

---

Model Architecture

The CNN model (Neural_Net) is built using PyTorch nn.Module.

🔹 Layers:

• Conv Layer 1

  • Input: 3 channels
  • Output: 12 channels
  • Kernel: 5×5

• Max Pooling

  • Kernel: 2×2
  • Stride: 2

• Conv Layer 2

  • Input: 12 channels
  • Output: 24 channels
  • Kernel: 5×5

• Fully Connected Layers

  • FC1: 24 × 5 × 5 → 120
  • FC2: 120 → 84
  • FC3 (Output): 84 → 10

Activation Function: ReLU (used after conv + FC layers)

---

Training Details

• Loss Function: CrossEntropyLoss
• Optimizer: SGD
  • Learning Rate: 0.001
  • Momentum: 0.9
• Batch Size: 32
• Epochs: 30

---

Results

After training for 30 epochs, the model achieved:

👉 Accuracy: 68.73% on CIFAR-10 test dataset

(Note: This is a simple baseline project. You can increase the accuracy by adding data augmentation, increasing the number of convolutional layers, adding Dropout/Batch Normalization, or training for more epochs!)

---

Custom Image Inference

The notebook supports testing on your own images.

Preprocessing:

• Resize to 32×32
• Convert to tensor
• Normalize

Example Predictions:

Image	Description	Prediction
11.png	Cat image	cat
22.png	Cartoon ship	ship
33.png	Cruise ship	ship
image.png	Airplane	plane

---

How to Run

1. Clone this repository:

git clone https://github.com/imrancoder786/image_classification

2. Open the notebook:

image_classfication.ipynb

3. Run all cells:

• Dataset download
• Model training
• Evaluation

---

Using Pretrained Model (Skip Training)

If you want to skip training:

• Load the pretrained model:

model.load_state_dict(torch.load("first_classification_model.pth"))

• Run evaluation or inference directly

---

Key Highlights

$ Simple CNN architecture
$ End-to-end pipeline (training → evaluation → inference)
$ Custom image prediction support
$ Beginner-friendly deep learning project