MeshMAE: Masked Autoencoders for 3D Mesh Data Analysis

Recently, self-supervised pre-training has advanced Vision Transformers on various tasks w.r.t. different data modalities, e.g., image and 3D point cloud data. In this paper, we explore learning paradigm for mesh analysis based Transformers. Since applying Transformer architectures to new modalities is usually non-trivial, first adapt processing, i.e., Mesh Transformer. specific, divide a into several non-overlapping local patches with each containing the same number of faces use position patch’s center form positional embeddings. Inspired by MAE, how Transformer-based structure benefits downstream tasks. We randomly mask some feed corrupted Then, through reconstructing information masked patches, network capable discriminative representations Therefore, name our method MeshMAE, which can yield state-of-the-art or comparable performance tasks, classification segmentation. addition, also conduct comprehensive ablation studies show effectiveness key designs in method.