T-CorresNet: Template Guided 3D Point Cloud Completion with Correspondence Pooling Query Generation Strategy

TL;DR

This work tackles the challenge of completing incomplete 3D point clouds by introducing T-CorresNet, which combines a Gaussian spherical template with a Corres-Pooling query generator to guide a coarse-to-fine completion pipeline. The encoder embeds a spherical prior into the input, producing a coarse template, while the Corres-Pooling module refines this template and generates dynamic query tokens for a Transformer-based decoder, culminating in FoldingNet-based final reconstruction. The approach achieves state-of-the-art or competitive results on PCN and ShapeNet-55/34 benchmarks, with ablations confirming the positive contributions of both the spherical template guidance and the Corres-Pooling strategy. This work enhances global shape priors and detail preservation in point cloud completion and shows strong generalization to unseen categories, underlining its practical impact for autonomous systems and manufacturing. The code is publicly available, enabling reproducibility and further research into template-guided geometric completion.

Abstract

Point clouds are commonly used in various practical applications such as autonomous driving and the manufacturing industry. However, these point clouds often suffer from incompleteness due to limited perspectives, scanner resolution and occlusion. Therefore the prediction of missing parts performs a crucial task. In this paper, we propose a novel method for point cloud completion. We utilize a spherical template to guide the generation of the coarse complete template and generate the dynamic query tokens through a correspondence pooling (Corres-Pooling) query generator. Specifically, we first generate the coarse complete template by embedding a Gaussian spherical template into the partial input and transforming the template to best match the input. Then we use the Corres-Pooling query generator to refine the coarse template and generate dynamic query tokens which could be used to predict the complete point proxies. Finally, we generate the complete point cloud with a FoldingNet following the coarse-to-fine paradigm, according to the fine template and the predicted point proxies. Experimental results demonstrate that our T-CorresNet outperforms the state-of-the-art methods on several benchmarks. Our Codes are available at https://github.com/df-boy/T-CorresNet.

Figures (3)

• Figure 1: An overview of the proposed T-CorresNet. We combine the transformer encoder-decoder architecture with the spherical template, following the coarse-to-fine generation paradigm. Our encoder module mainly predicts the global point proxies and generates the coarse complete template model under the guide of the spherical prior. Then through the Corres-Pooling query generator, we fetch the dynamic query tokens along with the value vectors, which could predict the accurate complete point proxies by the transformer decoder module. The FoldingNet could generate the complete results based on the fine template and the predicted point proxies.
• Figure 2: The pipeline of Corres Attention module.
• Figure 3: Visualization results on PCN dataset.