A Fine-Grained Clothing Style Image Segmentation Method Based on DeepLab v3+

Currently, research on artificial intelligence in the field of clothing production and manufacturing is still lacking. To advance the process of AI in clothing production and manufacturing, it is necessary to solve the problem of fine-grained segmentation of clothing style images to extract useful information. This paper first constructs a new dataset based on clothing structure morphology, manufacturing processes, and functional characteristics—the Clothing Structure Attribute (CSA) Dataset. Then, based on the DeepLab v3+ model, we propose a clothing style image segmentation algorithm named Clothing Structure Attribute DeepLab v3+ (CSA-DeepLab v3+). In this algorithm, we add feature extraction layers to the baseline network architecture, modify the ASPP convolutional kernel structure, and use two attention mechanisms: CBAM and ECA. We also conduct comparative and ablation experiments to verify performance of our model on the dataset presented in this paper providing a new approach for the automation of clothing production and manufacturing.