Table 2 Statistical Table of Cultural Compliance and Semantic Evaluation of Shape Grammar Generation under Different Control Strategies.

From: A semantic reconstruction and AI-controlled generation method for the cultural genes of Qing Dynasty embroidery patterns: a case study of official rank badges

Control Type

Metric Description

Mean Score

Score Range

Performance Characteristics

Hard Constraint

CLIP-based Semantic Matching Score

0.06

0.03–0.07

Sparse distribution, abstract expression.

Geometric Matching Score (e.g., IoU)

0.07

0.02–0.07

High precision, concentrated score range.

Soft Constraint

CLIP-based Semantic Matching Score

0.25

0.12–0.35+

Broad distribution, significantly better than hard constraint.

Layout (Composition) Score

0.01

Extremely low

Lacks structural control capability.

Stitching Compliance Score

0.40

0.10–0.80

Flexible expression, rich in detail.

Performance by Stitch Type (Pan embroidery highest, padded stitch lower)

0.33–0.15

Varies by type

Complex structural stitches perform better.

Overall Recommendation

Suggested Application Scenarios for Control Strategies

–

–

Hard constraint is suitable for geometric fidelity, soft constraint is preferable for artistic expression.

  1. Statistically compares key performance indicators for the cultural compliance and semantic evaluation of images generated based on Shape Grammar under two different control strategies (Hard Constraints and Soft Constraints), including mean score, score range, corresponding performance characteristics, and application suggestions.
Back to article page