Floor PLanning Using Gen-AI is an AI-driven project developed by Team ThinkML during the Neurathon hackathon at NIT Silchar, where it secured the 2nd prize. The project focuses on generating detailed and accurate floor plans using a combination of fine-tuned models and pre-trained multi-modal systems.
- AI-Generated Floor Plans: Leverages advanced AI models to create precise floor plans based on user inputs.
- Model Fine-Tuning: Utilizes fine-tuned versions of Stable Diffusion with ControlNet to enhance output control and accuracy.
- API Integrations: Incorporates several pre-trained models via APIs, including:
- Janus 1.3B and Janus Pro: Advanced text-to-image generation capabilities.
- Optimized_Cloudqi_Text_to_Image_Model and Cloudqi_Text_to_Image_Model: Enhanced image generation performance.
- Vertex AI: Cloud-based AI tools for scalable model inference.
- Custom Safety Mechanisms: Implements NSFW filtering and addresses common errors in ControlNet-based pipelines.
- Performance Optimization: Configures the pipeline for efficient GPU utilization and memory management, ensuring faster and more reliable image generation.
- Models/: Contains the fine-tuned models used for floor plan generation.
- Notebooks/: COlab notebooks demonstrating the model training and inference processes.
- Results/: Sample outputs and results from the model.
- Testing/: Scripts and data for testing the model's performance and accuracy.
- backend/: Backend code handling API calls and data processing.
- src/: Source code for the frontend interface and integration with the backend.
- Python 3.x
- PyTorch
- Transformers
- Diffusers
- PIL
- NumPy
Download the repo and run the notebooks with run time as T4 GPU.
Contributions are welcome! Please fork the repository and submit a pull request with your changes.
This project is licensed under the MIT License. See the LICENSE file for details.
We extend our gratitude to NIT Silchar and their Machine Learning Club for organizing Neurathon and providing a platform for innovation.
For more details, visit our GitHub repository: Neurathon-25
Developed with ❤️ by Team ThinkML