Improving the Performance of Industrial Dyeing Recipe Models by Taking Log-Transform in the Data Pre-Processing

Abstract

Recent research advances on fabric dyeing have focused on modeling the relationship between dye concentrations and the final color on fabrics. The emerging techniques in related studies have great potential to evolve the traditional dyeing industry to manufacture much more smartly. Given that dyeing is a complex process regulated by many factors, one of the challenging problems in the aforementioned techniques is to maintain the modeling accuracy at acceptable level. Other than developing high-performance algorithms and model architectures, it is also important to include effective data pre-processing techniques in modeling. In this paper, we show that conducting log-transform to the industrial dyeing data can greatly improve the performance of industrial dyeing recipe models. Such observations are confirmed on modeling tasks using different formats of color as input and different types of loss function in the model training. These findings may provide useful implications for related studies for the dyeing industry.