Object Recognition Drawing Program

Overview

This project is a simple drawing application that allows users to draw objects and get them recognized using a pre-trained Convolutional Neural Network (CNN) model. The project is divided into two main parts: the drawing application using Pygame and the model training using TensorFlow/Keras.

Features

1. Drawing Application:

   • Allows users to draw objects on a 500x500 pixel canvas.
   • Recognizes the drawn object and displays the name of the object along with the confidence level.
   • Uses Pygame for creating the drawing interface.

2. Model Training:

   • Loads Quick, Draw! dataset for training.
   • Preprocesses the data and trains a CNN model to recognize different categories of objects.
   • Uses data augmentation to improve model generalization.
   • Saves the best model based on validation loss.

Dependencies

• Python 3.6 or later
• Pygame
• NumPy
• TensorFlow/Keras

How to Run

1. Install dependencies: Ensure you have all the necessary libraries installed. You can install them using pip:

   pip install pygame numpy tensorflow

2. Train the Model:

   • Place the Quick, Draw! dataset files in the specified directory.
   • Run the model training script to train and save the model.

3. Run the Drawing Application:

   • Execute the Pygame script to start the drawing application.
   • Draw an object on the canvas and release the mouse button to see the prediction.

Model Training Details

• The model is trained on the Quick, Draw! dataset which consists of various categories of doodles.
• The dataset is preprocessed by normalizing the pixel values and reshaping the images.
• A CNN model is used for training with layers including Conv2D, MaxPooling2D, Flatten, Dense, and Dropout.
• Data augmentation techniques such as rotation, zoom, and shift are applied during training.
• The model is compiled using the Adam optimizer and categorical cross-entropy loss function.
• Model training includes callbacks for saving the best model and reducing the learning rate when the model plateaus.

Drawing Application Details

• The drawing application is built using Pygame.
• Users can draw on the canvas using the mouse. The drawing is converted to a grayscale image and resized to 28x28 pixels.
• The pre-trained model is used to predict the category of the drawn object.
• The recognized object name and confidence level are displayed on the screen.

Conclusion

This project provides a basic framework for creating a drawing application with object recognition capabilities. The model can be trained on different categories of objects, and the application can be extended with additional features and improvements.

License

This project is licensed under the MIT License - see the LICENSE file for details.