Computational study demonstrates high grading accuracy for student artwork using combined AI algorithms, suggesting effective auxiliary feedback in creative education.
Key Points
To construct an automated art design teaching evaluation system that simultaneously assesses image-text semantic concepts and technical visual quality.
Combined Contrastive Language-Image Pre-training (CLIP) for semantic feature extraction (theme, style, design intention) with Spatial Feature Image Quality Assessment (SF-IQA) for no-reference technical quality scoring.
Developed an adaptive feature-level fusion module to integrate semantic and technical evaluations into a unified scoring metric.
Achieved matching degrees of 89.7% for color coordination, 87.4% for composition logic, and 91.2% for style consistency.
Demonstrated a Pearson correlation of 0.91 with teacher grading benchmarks, a mean absolute error of 3.2 points, and an average runtime of 2.1 minutes per artwork (a 5.71-fold speed-up over manual evaluation).