Hyper-models-in-computer-vision-

Using computer vision Hyper-models on Chest X-Ray Images (Pneumonia)

🩻 Pneumonia Detection from Chest X-Rays using EfficientNet (Baseline & HyperModel)

📌 Project Overview

This project focuses on detecting Pneumonia from Chest X-Ray images using deep learning techniques. The dataset used is the publicly available Chest X-Ray Pneumonia dataset from Kaggle, which includes images categorized into train, val, and test folders.

The project is divided into two main parts:

• A baseline model using EfficientNet-B0
• A HyperModel using EfficientNet-B3 with advanced training enhancements

---

✅ Dataset Description

• Source: Paul Mooney – Chest X-Ray Pneumonia (Kaggle link)
• Classes: NORMAL, PNEUMONIA
• Structure:
  • train/: Used for model training
  • val/: Used for validation during training
  • test/: Used for final model evaluation

---

🧠 Part 1: Baseline Model with EfficientNet-B0

The first model uses a lightweight and efficient CNN, EfficientNet-B0, to classify chest X-ray images. It was trained using standard data preprocessing techniques without any complex optimization strategies.

Key Details:

• Input size: 224x224
• Optimizer: Adam
• Loss: CrossEntropyLoss
• Trained for 5 epochs
• Evaluated using the separate test dataset

🔍 Result:
The baseline EfficientNet-B0 model achieved an average accuracy of 81.09% on the test dataset.

---

🚀 Part 2: HyperModel with EfficientNet-B3

To improve performance, a more advanced version was built using EfficientNet-B3, paired with a range of optimization strategies:

Enhancements Applied:

• Advanced data augmentation (random flip, rotation)
• Learning rate scheduler for adaptive training
• Early stopping to prevent overfitting
• Confusion matrix and classification report for detailed test evaluation
• EfficientNet-B3 backbone for improved feature extraction

📈 Result:
The HyperModel achieved higher accuracy than the baseline model, outperforming the initial 81.09% by a significant margin, while also providing deeper insight into the model's performance across both classes.

---

📊 Example Output

Epoch 5: Train Loss=0.2341, Val Loss=0.1890, Val Acc=92.65%

Classification Report:
              precision    recall  f1-score   support
NORMAL           0.91       0.94      0.92       234
PNEUMONIA        0.97       0.94      0.95       390