SDXL model-based optimization for interior design: Data-driven and deep learning methods.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Computational Design

Background:

• AI-assisted interior design faces challenges with structural consistency and aesthetic fidelity.
• Generic diffusion models like Stable Diffusion XL (SDXL) require domain-specific adaptation for spatial design tasks.

Purpose of the Study:

• To propose a novel, domain-specific optimization framework for SDXL tailored to interior design.
• To enhance structural consistency and aesthetic fidelity in AI-generated interior designs.

Main Methods:

• Developed a systematic pipeline integrating automated semantic cleaning and hyperparameter optimization.
• Constructed a high-quality, annotated dataset using a semi-automated YOLO-based filtering process.
• Established an empirically validated training protocol with optimal Dropout, L1/L2 regularization, and dynamic learning rates.

Main Results:

• The optimized framework significantly outperformed baseline models in Fréchet Inception Distance (FID), Structural Similarity Index (SSIM), and Learned Perceptual Image Patch Similarity (LPIPS).
• Achieved robust CLIP Semantic Alignment, indicating improved understanding of design concepts.
• An ablation study confirmed semantic cleaning and structural regularization are critical for geometric fidelity, reducing FID by 51.1%.

Conclusions:

• The proposed framework offers a technically robust methodology for adapting large-scale diffusion models to specialized spatial design requirements.
• This research advances AI capabilities in creating structurally consistent and aesthetically pleasing interior designs.