Phased Digital Garment Pattern Arrangement on 3D Avatars

The patent segments the clothing patterns into multiple batches and processes them in sequential iterations rather than simultaneously. Each iteration processes a subset of patterns that are arranged on the 3D avatar, with subsequent iterations processing remaining patterns. This segmentation reduces the computational complexity of each iteration while ensuring all patterns are eventually processed to achieve complete garment generation.

The patent performs preliminary actions by determining sewing relationships between all clothing patterns before the iterative arrangement process begins. The sewing relationships, which define how patterns should be connected, are pre-calculated and stored. During each iteration, these pre-determined relationships guide the arrangement process, reducing the complexity of each iteration step while maintaining high arrangement accuracy through multiple passes.

The patent implements self-service by training the machine learning body part determination model offline beforehand. The trained model is then deployed for inference during the pattern arrangement process. This approach requires significant computational energy only during the one-time training phase, while the actual garment generation process uses the pre-trained model for efficient, low-energy automated body part determination, reducing real-time computational resource requirements.

The patent applies local quality by performing intersection removal selectively and locally on only those clothing patterns that exhibit intersections after alignment, rather than processing all patterns uniformly. The system identifies specific patterns with intersection issues and applies correction algorithms only to those local cases, preserving visual quality where needed while minimizing unnecessary processing time for patterns that are already correctly positioned.