This image classification project repository, where I explore how Convolutional Neural Networks (CNNs) can empower machines to see and recognize patterns in visual data.
This project demonstrates my ability to design, train, and evaluate a custom CNN for classifying images using popular datasets. It was developed as part of my practical coursework and self-learning.
To build an end-to-end pipeline that:
- Preprocesses image data for training,
- Designs a CNN from scratch using Keras,
- Trains the model efficiently with proper regularization,
- Evaluates model performance on unseen data,
- Visualizes metrics and interprets learning curves.
- ✅ Built a custom Convolutional Neural Network from scratch
- ✅ Achieved high accuracy on image classification tasks
- ✅ Applied techniques like dropout, data augmentation, and early stopping
- ✅ Used TensorFlow/Keras for efficient model development
- ✅ Visualized confusion matrices and training performance
- Language: Python
- Frameworks: TensorFlow, Keras
- Libraries: NumPy, Matplotlib, Seaborn, Scikit-learn
- Tools: Jupyter Notebook, Google Colab
- Source: Publicly available datasets such as CIFAR-10 / Fashion MNIST
- Type: Multi-class image classification (e.g., clothes, animals, objects)
- Size: Thousands of labeled images in grayscale or RGB format
- Trained with batch normalization and dropout to reduce overfitting
- Achieved >85% accuracy (depending on dataset)
- Plotted learning curves to assess convergence
- Evaluated with classification reports & confusion matrices
- 📧 Email: hirdeshpal15@gmail.com
- 🔗 LinkedIn – Hirdesh Pal
