Unsupervised Domain Adaptation Method, Apparatus, System and Storage Medium for Semantic Segmentation Based on Uniform Clustering

Abstract

The invention discloses an unsupervised domain adaptation method, device, system and storage medium for semantic segmentation based on uniform clustering. First, a prototype-based source domain uniform clustering loss and an empirical prototype-based target domain uniform clustering loss are established. Loss, which reduces the intra-class difference of pixels of the same category, and drives pixels with similar structure but different categories to move away from each other, tends to be evenly distributed, increases the inter-class distance, and overcomes the problem of unclear category boundaries in the process of domain adaptation; then, Integrating the prototype-based uniform clustering loss in the source domain and the target domain uniform clustering loss based on the empirical prototype into the adversarial training framework, it narrows the domain difference between the source domain and the target domain, and enhances the cross-domain data on the semantic segmentation model. Adaptability, thereby improving the accuracy of semantic segmentation.

Description

technical field [0001] The invention belongs to the field of unsupervised domain adaptation, and in particular relates to an unsupervised domain adaptation method, device, system and storage medium based on uniform clustering semantic segmentation. Background technique [0002] In the past few decades, semantic segmentation models based on convolutional neural networks have achieved remarkable success relying on large-scale labeled datasets and have shown great potential in areas such as autonomous driving and robotics. However, these large-scale labeled data sets require long-term intensive manual labeling by technicians with professional knowledge, which consumes expensive labor and time costs. [0003] To alleviate this problem, recent research efforts have gradually attempted synthetic datasets, such as: GTA5, SYNTHIA, and Synscapes. Synthetic datasets can be automatically synthesized by game engines or simulators, reducing labor costs. However, these models trained on...

AI Technical Summary

Problems solved by technology

However, these indirect strategies still do not overcome the problem of unclear class structure

On the other hand, although most clustering-based unsupervised domain adaptation methods adjust the category structure, they are mainly applied to image-level classification tasks without considering the high-dimensional semantic feature space, resulting in limited ability to identify semantic information.

Images