Instance and prototype contrastive learning for multi-view 3D model retrieval and classification

Yaqian Zhou

doi:10.1016/j.ipm.2026.104655

Instance and prototype contrastive learning for multi-view 3D model retrieval and classification

2026-01-28

0

PRE

AI

DOI：10.1016/j.ipm.2026.104655

Original

Original request for help

Save

Abstract

En 中文

• We propose an Instance and Prototype Contrastive Learning (IPCL) framework for unsupervised multi-view 3D model representation learning, which significantly enhances local salience and global discriminative semantic representation by exploring multi-view and class contexts. • We design a view-level instance contrastive learning module that selects meaningful contrastive samples to refine locally salient and invariant visual features of 3D models. • We design a prototype contrastive learning module based on the Wasserstein selection to enhance the discriminative global representation of 3D models and mine representative class prototypes. • Extensive experimental results on two challenge datasets verify the effectiveness of our method and achieve state-of-the-art results in 3D model retrieval and classification tasks.

Instance and prototype contrastive learning for multi-view 3D model retrieval and classification

Abstract

Journal

Organization