Abstract
Analysis of the 3D Texture is indispensable for various tasks, such asretrieval, segmentation, classification, and inspection of sculptures, knittedfabrics, and biological tissues. A 3D texture is a locally repeated surfacevariation independent of the surface's overall shape and can be determinedusing the local neighborhood and its characteristics. Existing techniquestypically employ computer vision techniques that analyze a 3D mesh globally,derive features, and then utilize the obtained features for retrieval orclassification. Several traditional and learning-based methods exist in theliterature, however, only a few are on 3D texture, and nothing yet, to the bestof our knowledge, on the unsupervised schemes. This paper presents an originalframework for the unsupervised segmentation of the 3D texture on the meshmanifold. We approach this problem as binary surface segmentation, partitioningthe mesh surface into textured and non-textured regions without priorannotation. We devise a mutual transformer-based system comprising a labelgenerator and a cleaner. The two models take geometric image representations ofthe surface mesh facets and label them as texture or non-texture across aniterative mutual learning scheme. Extensive experiments on three publiclyavailable datasets with diverse texture patterns demonstrate that the proposedframework outperforms standard and SOTA unsupervised techniques and competesreasonably with supervised methods.