Monument-and-Architecture-Detection

Overview

This project focuses on recognizing landmarks and monuments using deep learning models. It includes two distinct implementations:

• Architecture.py: Uses a pre-trained MobileNetV2 model for predicting prominent landmarks based on images provided by the user.

• Monument_Prediction.py: Builds and trains a custom Convolutional Neural Network (CNN) model to classify monuments into predefined categories using a dataset of images.

Features

1. Architecture.py

• Leverages the MobileNetV2 model pre-trained on ImageNet.

• Preprocesses input images for prediction.

• Displays top-5 predicted classes for the input image.

• Visualizes the input image using OpenCV.

2. Monument_Prediction.py

• Implements a custom CNN architecture.

• Provides utilities for data preprocessing, visualization, and training.

• Trains on a dataset using TensorFlow and Keras.

• Saves the trained model for future use.

• Offers functionalities to:

View random images from the dataset.

• Plot training and validation loss/accuracy curves.

• Predict and visualize results for new images.

Requirements

1. Python 3.7+

2. Libraries:

• TensorFlow

• OpenCV

• NumPy

• Pandas

• Matplotlib

• Google Colab (for cloud-based training)

Installation

1. Clone the repository

git clone https://github.com/your_username/landmark-monument-recognition.git

2. Navigate to the project directory

cd landmark-monument-recognition

3. Install required libraries

pip install -r requirements.txt

Usage

1. Architecture.py

• Provide the path to the image you want to analyze in the image_path variable.

• Run the script:

python Architecture.py

• View the top-5 predictions along with the image.

2. Monument_Prediction.py

• Prepare the dataset:

Place training images in train/ directory.

Place testing images in test/ directory.

Update the dataset path in the path variable.

• Train the model:

python Monument_Prediction.py

• Save the trained model for future use.

• Test new images using the pred_and_plot function.

Model Training and Evaluation

1. CNN Architecture

• The simplified CNN architecture includes:

• Convolutional layers with ReLU activation.

• MaxPooling layers for down-sampling.

• A Flatten layer to convert feature maps to a single vector.

• Dense layers with softmax activation for multi-class classification.

2. Training

• The model is trained using categorical cross-entropy loss and the Adam optimizer.

• Accuracy and loss metrics are logged for both training and validation.

3. Visualization

• Loss and accuracy curves are plotted using the plot_loss_curves function.

Sample Results

• Run pred_and_plot() to test the trained model on new images.

ex: pred_and_plot(model_10, "path_to_image.jpg", class_names)

Saving and Loading Models

The trained model is saved in TensorFlow format:

model_10.save("saved_trained_model")

Additionally, the model is saved in a pickle file for easier distribution:

with open('saved_trained_model.pkl', 'wb') as file: pickle.dump(model_10, file)

Dataset

The dataset should be organized as:

dataset/

├── train/

│ ├── class1/

│ ├── class2/

│ └── ...

├── test/

| ├── class1/

| ├── class2/

└── ...

• Dataset Download

Kaggle Landmark Recognition Datasets

Google Open Images Dataset

Update your dataset path accordingly before training the model.

Future Improvements

• Expand the dataset for more diverse classes.

• Experiment with advanced pre-trained models (e.g., EfficientNet, ResNet).

• Optimize hyperparameters for better performance.

• Add functionality for automated dataset preprocessing.

Acknowledgments

• Pre-trained MobileNetV2 model: TensorFlow Keras Applications.

• Data Augmentation and Training Utilities: TensorFlow and Keras libraries.

• Visualization: Matplotlib.