Semi‐supervised Mesh Segmentation and Labeling

Abstract

Recently, approaches have been put forward that focus on the recognition of mesh semantic meanings. These methods usually need prior knowledge learned from training dataset, but when the size of the training dataset is small, or the meshes are too complex, the segmentation performance will be greatly effected. This paper introduces an approach to the semantic mesh segmentation and labeling which incorporates knowledge imparted by both segmented, labeled meshes, and unsegmented, unlabeled meshes. A Conditional Random Fields (CRF) based objective function measuring the consistency of labels and faces, labels of neighbouring faces is proposed. To implant the information from the unlabeled meshes, we add an unlabeled conditional entropy into the objective function. With the entropy, the objective function is not convex and hard to optimize, so we modify the Virtual Evidence Boosting (VEB) to solve the semi-supervised problem efficiently. Our approach yields better results than those methods which only use limited labeled meshes, especially when many unlabeled meshes exist. The approach reduces the overall system cost as well as the human labelling cost required during training. We also show that combining knowledge from labeled and unlabeled meshes outperforms using either type of meshes alone.

A data-driven editing framework for automatic 3D garment modeling

“…We also make comparisons to both the supervised method of [27] and the semi-supervised approach [35]. The robustness of our method to inconsistently labeled training meshes, such as human mislabeling and label missing, is also illustrated.…”

“…Recently, Wang et al [56] proposed a semi-supervised learning method to improve the quality of segmentations among a shape collection using a sparse set. Huang et al [35] developed semi-supervised mesh segmentation and labeling to generate better results than both supervised and unsupervised method. And these methods overcome the difficulties of requiring a large amount of labeled meshes and the inability to use unlabeled meshes.…”

“…Compared with other rigid 3D models, the garment models in our library typically have the prototype categories and detailed attributes. In our implementation, we use the semi-supervised segmentation and labeling technique of Huang et al [35], assisted by manual segmentation and labeling. The method uses information from both labeled and unlabeled meshes to yield better results, overcoming the difficulties of requiring a large amount of labeled meshes and the disadvantage of inability to use unlabeled meshes.…”

Data-driven shape analysis and processing

“…The parameters θ of Equation 2 can be learned not only from the training labeled shapes, but also from the unlabeled shapes [LCHB12]. The idea is that learning should maximize the likelihood function of the parameters over the labeled shapes, and also minimize the entropy (uncertainty) of the classifier over the unlabeled shapes (or correspondingly maximize the negative entropy).…”

A Hybrid Boosted-SVM Classifier for Recognizing Parts of 3D Objects

“…Therefore, we aim to strengthen the classifier used for recognition. There exist also other systems for recognizing components of 3D objects; the majority of them used Adaboost algorithm variants to build functions used for recognition [14,[15][16]. Most of these proposed approaches using machine learning used a multitude of geometric and topologic characteristics and need a post-processing to refine the obtained results.…”