Room-Layout-Estimation-in-AR-MR2024

This repository provides a streamlined pipeline for room layout estimation in AR, optimized for Magic Leap 2. By integrating lightweight 2D boundary segmentation (ST-RoomNet), refined 2D and 3D line generation, and user-centric rendering, the system delivers robust performance and intuitive interaction. Clutter is reduced and structural clarity is enhanced, resulting in a comfortable, user-friendly experience. Future directions include exploring long-range depth sensing, advanced room layout models, and improved AR interfaces.

1. Environment Setup

1. Clone this repository to your local machine.
2. Install dependencies with:

   pip install -r requirements.txt
   

2. Testing the Backend Part Locally

1. Open and run test.ipynb to see the entire process demonstrated with an example dataset.
2. This notebook walks through:
   • Initializing the environment
   • Loading images and metadata
   • Estimating layouts
   • Rendering results for inspection

3. Running on Magic Leap 2

For real-world usage on Magic Leap 2, follow these steps:

1. Start the Monitoring Script
   Run run_monitor.bat to begin monitoring for new images. This script continuously checks for fresh data coming from the Magic Leap 2 device.

2. Run the Data-Capture Script
   Execute newtxt.py in parallel. This script handles capturing new images from the device.

3. Automatic Indexing
   Each newly captured image is automatically assigned an index in index.json located under the received_data folder, alongside:

   • The color image
   • The depth image
   • Associated metadata

4. Rendering
   Once the new images and metadata are received, the room layout estimation and rendering pipeline will process them for visualization.

4. Testing the Workflow Locally

1. Run run_monitor.bat and newtxt.py as explained above.
2. Manually add an index in index.json.
3. Important Note: Because Unity uses a left-hand coordinate convention, there are slight differences in how lines are rendered here (for real time case) compared to the methodes shown in Section 2 (for verification case, all right-hand assumptions used). Nonetheless, the final outcomes have been verified to be correct during real-world rendering.

Enjoy exploring room layout estimation for AR/MR applications!