AI-Based interior design editor
An algorithm enabling rapid room interior editing based on a single photo: changing floor and wall materials, adding or removing furniture, while preserving accurate scene geometry and lighting shadows.
Tasks
- Develop an algorithm for segmenting key surfaces and interior objects (floor, walls, ceiling, furniture).
- Reconstruct room layout and camera parameters (position, field of view, vanishing points).
- Implement overlay of new materials on segmented surfaces while accounting for perspective.
- Enable insertion and removal of 3D furniture models.
- Transfer original shadows to new materials to maintain scene realism.
- Create a prototype web application for real-time room interior editing based on a photo.
- Integrate and train segmentation models (SWIN, FCN, HRNET, ANN) for precise surface detection.
About the Project
The project’s goal is to provide users with a web tool that allows them to upload a room photo and instantly “try on” new finishing materials or furniture arrangements.
This required reconstructing the 3D geometry of the room from a single image, determining the camera position, segmenting main surfaces (floor, walls, ceiling) and existing objects, and then accurately applying new textures and shadows.
Results
Challenges and Solutions
The primary challenge was reconstructing an accurate 3D scene from a single photo. Rooms may have non-standard shapes, varied lighting, and partially occluded surfaces. To address this:
- Combination of multiple segmentation networks improved mask quality and enabled flexible model selection based on the scene.
- Hybrid layout reconstruction (LSUN-Room / NonCuboidRoom) combined with vanishing point detection enabled accurate angle and perspective calculation, even for non-rectangular rooms.
- Shadow transfer was implemented using OpenCV: original shadows are extracted, adapted to new materials, and overlaid to maintain photorealistic results.
- Object removal with LaMa allows users to “clear” space before adding new furniture, reducing artifacts.
- Camera calibration via fSpy ensures accurate placement of 3D models in the scene, matching the original photo’s perspective.
Related Services
Artificial Intelligence
We create innovative solutions based on artificial intelligence for your business.
Read More
Outsourcing
Our team of skilled developers is ready to create websites and applications of any complexity. We build modern web solutions using cutting-edge technologies and architectural approaches.
Read MoreReady to discuss your project?
Describe your task, we will make a research and respond to you as soon as possible.