Exploring the applicability of pix2pix for firing effect prediction of Jingdezhen high-temperature colored-glaze porcelain panels
Year: 2026
Type: article
Abstract: High-temperature colored-glaze ceramics are distinguished by their kiln-transformation, glaze-transformation, and flow effects. However, complex stochastic firing processes make their final appearance difficult to predict, hindering craft transmission and industrial development. This study investigated a pix2pix-based simulation framework for estimating the fired appearance of Jingdezhen high-temperature colored-glaze porcelain panels. Panels produced with fixed glaze combinations were represented by semantic segmentation maps as conditional inputs. VGG perceptual loss and instance-noise regularization were incorporated into pix2pix. Performance was assessed using five panel-level holdout validation splits, SSIM, LPIPS, and color-histogram correlation. The baseline model generated relatively realistic fired appearances within the training domain. Among combined-constraint configurations, L1 = 30, perc = 10, and noise enabled achieved the lowest LPIPS (0.2011) and a relatively high SSIM (0.5229), while suppressing some GAN texture artifacts. Nevertheless, performance deteriorated substantially on unseen samples, and generated images remained far below practical requirements. These findings indicate that pix2pix’s deterministic mapping struggles with the complex stochastic relationships involved in ceramic firing. Future work should investigate multimodal generative architectures integrated with materials-mechanism modeling to support reliable tools for craft transmission and industrial development.
Source: Scientific Reports
Authors: Yitian Liu, Jiaqing Guan, Jiren Zhang, Yang Ning
Institution: Wuhan University of Science and Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7235646349
