CNN Classifier

A multiclass image classification project built using Convolutional Neural Networks (CNN) and PyTorch.
This model aims to classify various heart diseases based on medical images.

---

Features

• Multi-class classification using deep CNN architecture
• Modular and configurable training system
• Automated logging and visualization
• Configurable preprocessing and dataset management

---

First-Time Setup

1. Python Environment

Use Python 3.11.
Create a virtual environment and install required packages:

python3.11 -m venv venv
source venv/bin/activate

---

2. Install PyTorch

Follow the official PyTorch Get Started Guide
to install PyTorch for your system.

NOTE for AMD GPU users:
Install PyTorch manually before other requirements using:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.3

---

3. Install Other Requirements

Once PyTorch is installed, install the remaining dependencies:

pip install -r requirements.txt

---

Configuration File

Create a configuration file named config.json in the project root (same directory as config.py).

Example config.json

{
    "dataset_path": "/home/user/cnn-classifier/data",
    "labels_file_path": "/home/user/cnn-classifier/data/labels.csv",
    "logs_path": "/home/user/cnn-classifier/logs",
    "save_model_path": "/home/user/cnn-classifier/model",
    "model_name": "tf_efficientnet_b0.in1k",
    "final_img_width": 224,
    "num_channels": 1,
    "hyperparameters": {
        "learning_rate": 0.001,
        "batch_size": 1,
        "total_epoch": 100,
        "save_every": 1,
        "weight_decay": 0.0001,
        "max_norm": 1,
        "drop_rate": 0.3,
        "num_blocks_to_unfreeze": 2,
        "scheduler_type": "ReduceLROnPlateau",
        "scheduler_params": {
            "mode": "max",
            "factor": 0.1,
            "patience": 2
        }
    },
    "classes": [
        "Cardiomegaly",
        "Emphysema",
        "Effusion",
        "Hernia",
        "Infiltration",
        "Mass",
        "Nodule",
        "Atelectasis",
        "Pneumothorax",
        "Pleural_Thickening",
        "Pneumonia",
        "Fibrosis",
        "Edema",
        "Consolidation"
    ]
}

---

Preprocessing

Before training, you must preprocess your dataset.

Preprocessing performs:

• Image resizing to the target width
• Format conversion to .pth tensors
• Automatic train/validation/test splitting

Run Preprocessing

python preprocessing.py <data_dir>

Note: The preprocessing script copies files instead of moving them.

---

Directory Structure

After preprocessing, your dataset will follow this structure:

/config.dataset_path
├── train/
├── val/
└── test/

Each directory will be automatically created based on paths provided in the configuration file.

Logging

Training logs and plots are automatically generated under:

/logs/

Metrics such as loss, accuracy, and AUC are plotted and saved as .png images for each epoch.

---

License

This project is released under the MIT License.
Feel free to use and modify it for educational and research purposes.