🖼️ Image Recognition Using Deep Learning

🚀 Project Overview

This project implements an Image Recognition Model using Convolutional Neural Networks (CNNs) in PyTorch. The goal is to classify both living and non-living objects such as humans, animals, vehicles, and everyday objects.

We initially experimented with the CIFAR-10 dataset but have now moved on to a custom real-world dataset for improved accuracy and relevance.

---

📂 Dataset Information

✅ Custom Dataset

We are currently using a custom dataset located under:

real_dataset/
├── train/
│   ├── animals/
│   ├── humans/
│   └── vehicles/
├── test/
│   ├── animals/
│   ├── humans/
│   └── vehicles/

Each subfolder contains images classified by category. The dataset supports class-wise balanced sampling and augmentation techniques like random crop, flip, rotation, jitter, and affine transformations.

---

🧠 Model Architecture: ImprovedCNN

The model is defined in train.py and includes:

• ✅ 4 Convolutional Layers with Batch Normalization and ReLU
• ✅ MaxPooling after each convolutional block
• ✅ Dropout for regularization
• ✅ 3 Fully Connected Layers for classification
• ✅ Trained using CrossEntropyLoss and the Adam optimizer

---

🔍 Feature Map Visualization

You can now visualize feature maps from any convolutional layer (conv1, conv2, conv3, conv4) using the utils.py script.

How to Use:

python src/utils.py

🔧 Set the correct image_path and model_path inside utils.py

📁 Output: Feature map visualizations will be saved as:

• feature_maps_conv1.png
• feature_maps_conv2.png
• feature_maps_conv3.png
• feature_maps_conv4.png

---

🧪 Evaluation & Insights

Training includes early stopping and learning rate scheduling

Model performance is evaluated using:

• 🔢 Class-wise accuracy
• 📉 Training & validation loss curves
• 🧩 Confusion matrix visualized as a heatmap

All metrics and plots are saved automatically for easy access

---

Steps to Clone and Run the Project

1. Clone the repository:

   git clone https://github.com/AkarshanGhosh/Image_recognisation.git
   cd Image_recognisation

2. Run the data_loader.py:

   python src/data_loader.py

3. Train the Model:

   python src/train.py

4. Visualize Feature Maps:

   python src/utils.py

5. Make Predictions:

   python src/predict.py 

---

📊 Results

• 🔥 High classification accuracy on custom dataset
• 🧠 Clear feature progression observed in feature maps from conv1 to conv4
• 📊 Clean visualizations of training loss and validation accuracy

---

👩‍💻 Contributors

• Akarshan Ghosh

---

📌 Future Work

• Add Grad-CAM based heatmap visualizations
• Create a Streamlit/Web UI for real-time predictions
• Extend dataset and support multi-label classification