Welcome to Dino_Dash_AI! This project showcases an AI-powered bot that plays the popular Chrome Dino Game. Leveraging PyTorch and EfficientNet, this bot can automatically navigate the game and avoid obstacles with precision.

• Overview
• Features
• Installation
• Usage
• Training the Model
• Implementation Details
• Results
• Contributing
• License
• Acknowledgments

Dino_Dash_AI uses a deep learning model trained with PyTorch to recognize obstacles in the Chrome Dino Game and make real-time decisions about when to jump. The project demonstrates the power of AI in gaming and provides insights into the training and implementation of an AI game bot.

• EfficientNet v2_s Architecture: High accuracy with a small number of parameters.
• Real-Time Decision Making: The bot makes decisions on the fly, ensuring smooth gameplay.
• Data Augmentation: Enhances training data to improve model generalization.
• Detailed Documentation: Comprehensive blog post explaining the training process, model architecture, and bot implementation.

• Python 3.8 or higher
• PyTorch
• OpenCV
• NumPy

1. Clone the repository:

   git clone https://github.com/Hairul343/Dino_Dash_AI.git
   cd Dino_Dash_AI

2. Install dependencies:

   pip install -r requirements.txt

1. Start the Chrome Dino Game: Open the Chrome browser and go to chrome://dino.

2. Run the bot:

   python dino_bot.py

The bot will start playing the game automatically.

To train the model from scratch:

1. Prepare the Dataset: Collect images from the Chrome Dino Game, labeling obstacles and actions (jump or no jump).

2. Train the Model:

   python train_model.py --data_path ./data --epochs 50 --batch_size 32

3. Evaluate the Model:

   python evaluate_model.py --model_path ./models/efficientnet_v2_s.pth --test_data_path ./test_data

The bot uses the EfficientNet v2_s architecture, chosen for its balance between accuracy and computational efficiency. The model is trained to recognize obstacles and decide when to jump.

Data augmentation techniques such as random cropping, flipping, and color jittering are used to increase the diversity of the training data and improve model generalization.

The bot captures the game screen in real-time, processes the image to detect obstacles, and sends jump commands based on the model's predictions.

The bot demonstrates high accuracy in recognizing obstacles and making timely jumps, effectively navigating the Chrome Dino Game. Detailed performance metrics and results are available in the blog post linked below.

We welcome contributions! If you'd like to contribute, please fork the repository and submit a pull request with your changes. Ensure all tests pass before submitting your PR.

• Follow the existing code style.
• Write clear and descriptive commit messages.
• Include tests for any new features or bug fixes.

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

• Thanks to the developers of PyTorch and EfficientNet for their powerful tools.
• Special thanks to the contributors of the Chrome Dino Game for the fun and challenging game.
• For more details on the project, check out the detailed blog post.

For any questions or issues, please open an issue on this repository or contact the maintainers.