Optimization design of interior space based on the two-stage deep learning network and Single sample-driven method.

Area of Science:

• Artificial Intelligence
• Computer Vision
• Computational Design

Background:

• Traditional interior design methods face challenges in timeliness and uniqueness.
• Existing digital design tools often lack intelligent optimization and personalization capabilities.

Purpose of the Study:

• To develop an optimized interior space design framework using a two-stage deep learning network.
• To enhance design efficiency, uniqueness, and user adaptability in interior design solutions.
• To introduce a single-sample-driven mechanism for personalized interior design.

Main Methods:

• A two-stage deep learning network combining a Transformer network for feature extraction and a diffusion model for generative optimization.
• Utilizing multi-dimensional features from input space images, including spatial layout, color, and furniture style.
• Validation on public datasets: InteriorNet, SUN RGB-D, NYU Depth V2, and ScanNet.

Main Results:

• Reduced design cycle by 40% and increased space utilization by 25% compared to traditional methods.
• Improved proportional and scale coordination by 20% and color matching scores by 30% through personalized design.
• Demonstrated enhanced design efficiency, innovativeness, and user adaptability.

Conclusions:

• The proposed AI-driven framework offers an efficient and intelligent solution for interior space design.
• The two-stage deep learning network effectively integrates feature extraction and generative optimization.
• This study presents a new technological paradigm for artificial intelligence in design fields.