A Benchmark and VLM-Based Method for Quantifying Novelty-Usefulness-Coherence in Industrial Product Form Design

Aiming at the subjective assessment of industrial design creativity, this paper presents a quantitative assessment framework that combines novelty, appropriateness, and consistency and establishes a benchmark for assessing the modeling creativity of multi-category products. With the help of human subjective scores from human-computer collaborative labeling, this study designed a “design critic” model based on a visual language model to enable automatic creativity assessment. The experiment showed that the Spearman correlation coefficient between the design critic model and human ratings in the cross-category test was 0.67, while the appropriateness dimension scored 0.74. In the context of small sample migration, the self-supervised pre-trained model maintained a correlation of 0.61 after fine-tuning with only 20 samples. After three rounds of optimization, the average user preference rate increased by about 24%, with a peak of 68%.