Image Content Classification Model with Streamlit Web Application

Objective

The objective of this project was to develop a machine learning model capable of classifying images into three categories: 'Violent', 'Adult Content', and 'Safe'. The model was to be integrated into a Streamlit web application to demonstrate its functionality in a live web interface.

Tools and Libraries Used

• Python: Main programming language used for the project.
• TensorFlow/Keras: Used for building and training the convolutional neural network (CNN).
• NumPy: For data manipulation and preprocessing.
• Matplotlib: For creating visualizations of the training process.
• Streamlit: For creating an interactive web app to showcase the model's capabilities.
• Pillow: For handling image processing.

Key Steps

1. Environment Setup

   • Installed necessary libraries: TensorFlow, NumPy, Matplotlib, Streamlit, and Pillow.

2. Data Collection and Preprocessing

   • Collected images from various sources, categorized them into 'Violent', 'Adult Content', and 'Safe'.
   • Used TensorFlow's ImageDataGenerator to resize, normalize, and augment images.
   • Split the data into training and validation sets with an 80-20 ratio.

3. Model Development and Training

   • Constructed a CNN using MobileNetV2 as the base model.
   • Added custom layers on top of MobileNetV2 for specific classification tasks.
   • Compiled the model with the Adam optimizer and categorical cross-entropy loss function.
   • Trained the model using early stopping and a learning rate scheduler to prevent overfitting.
   • Applied class weights to handle class imbalance.

4. Model Evaluation

   • Evaluated the model using validation data.
   • Visualized training and validation accuracy and loss over epochs to assess model performance.

5. Streamlit Integration

   • Developed a Streamlit web app allowing users to upload images and view classification results.
   • Displayed model accuracy, loss, and training history within the web app.

Model Performance

• Training Results:

  • Achieved a training accuracy of approximately 80% and a validation accuracy of approximately 50% after 11 epochs.
  • Validation loss and accuracy fluctuated, indicating possible overfitting, but were managed using data augmentation, dropout, and class weights.

• Evaluation Metrics:

  • Validation Loss: 1.6966
  • Validation Accuracy: 0.7132

Streamlit Web Application

The Streamlit web application provides an interactive interface for users to:

• Upload an image for classification.
• View the uploaded image and its predicted category.
• Display confidence scores for each category.
• Show model performance metrics such as validation loss and accuracy.
• Visualize training history for accuracy and loss if available.

Conclusion

The project successfully developed a machine learning model to classify images into 'Violent', 'Adult Content', and 'Safe' categories with a reasonable degree of accuracy. Integrating the model into a Streamlit web application provided an interactive platform to demonstrate the model's capabilities, allowing users to upload images and view real-time classification results.

Future work could focus on collecting more data, especially for underrepresented classes, and fine-tuning the model further to improve accuracy and reduce overfitting. Additionally, deploying the Streamlit app on a cloud platform would make it accessible to a broader audience.

Running the Streamlit App

To run the Streamlit app, use the following command:

streamlit run app.py

This will start the Streamlit server, and you can interact with the web app through your browser, uploading images and viewing classification results.

https://figshare.com/articles/dataset/Adult_content_dataset/13456484

https://www.kaggle.com/datasets/karandeep98/real-life-violence-and-nonviolence-data/data

https://www.kaggle.com/datasets/kartikeybartwal/graphical-violence-and-safe-images-dataset/code