Bird Classifier CNN

Led the development of a custom bird classification model in PyTorch, achieving 87% accuracy on a dataset of 3,113 bird images. Utilized data augmentation and GAN-based image generation techniques to reduce overfitting by 20%, enhancing model generalization. This project demonstrated proficiency in deep learning and image classification, addressing challenges in dataset variability.

Model Architecture

• Architecture Selection:
• Experimented with Variational Autoencoder (VAE) and Generative Adversarial Network (GAN).
• Chose a Convolutional Neural Network (CNN) due to its:
• Superior feature detection capabilities.
• Efficiency in training.
• Proven effectiveness in handling image data.

Sample Code of the Model

class BirdModel(nn.Module):
    def __init__(self, num_classes=20):
        super().__init__()
        # input = 224 x 224 x 3
        self.conv1 = nn.Conv2d(3, 10, 5)
        self.pool1 = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(10, 20, 5)
        self.pool2 = nn.MaxPool2d(2, 2)
        self.fc1 = nn.Linear(20 * 53 * 53, 50)
        self.fc2 = nn.Linear(50, 20)

    def forward(self, x):
        self.to(device)
        x = self.pool1(nn.functional.relu(self.conv1(x)))
        x = self.pool2(nn.functional.relu(self.conv2(x)))
        x = x.reshape(-1, 20 * 53 * 53)
        x = self.fc1(x)
        x = self.fc2(x)
        return x

Training Process

• Data Augmentation Techniques:
• Implemented:
• Image rotation.
• Color fixation changes.
• Random erasing.
• Horizontal and vertical flips.
• Designed to simulate diverse real-world scenarios and improve model robustness.
• Hyperparameter Tuning:
• Extensive tuning led to a 15% improvement in model precision and recall.
• Loss Functions:
• Optimized loss functions tailored for classification tasks to enhance performance.