🤖 AI Summary
To address inaccurate rigid transformation estimation in low-overlap point cloud registration (PCR), this paper proposes an adaptive multi-step refinement network. It employs a generalized unidirectional attention mechanism to dynamically focus on the overlapping regions estimated in the previous step; incorporates step-index conditional encoding to model iterative dependencies; and introduces multi-level transformation supervision with gradient-weighted training to enhance robustness under severe overlap scarcity. The architecture is fully end-to-end differentiable and represents the first effort to jointly integrate unidirectional attention and stepwise conditional modeling for PCR refinement. Evaluated on 3DMatch, 3DLoMatch, and KITTI benchmarks, the method achieves state-of-the-art performance: notably, an 80.4% recall on 3DLoMatch—an absolute improvement of 1.2% over prior work.
📝 Abstract
Point Cloud Registration (PCR) estimates the relative rigid transformation between two point clouds of the same scene. Despite significant progress with learning-based approaches, existing methods still face challenges when the overlapping region between the two point clouds is small. In this paper, we propose an adaptive multi-step refinement network that refines the registration quality at each step by leveraging the information from the preceding step. To achieve this, we introduce a training procedure and a refinement network. Firstly, to adapt the network to the current step, we utilize a generalized one-way attention mechanism, which prioritizes the last step's estimated overlapping region, and we condition the network on step indices. Secondly, instead of training the network to map either random transformations or a fixed pre-trained model's estimations to the ground truth, we train it on transformations with varying registration qualities, ranging from accurate to inaccurate, thereby enhancing the network's adaptiveness and robustness. Despite its conceptual simplicity, our method achieves state-of-the-art performance on both the 3DMatch/3DLoMatch and KITTI benchmarks. Notably, on 3DLoMatch, our method reaches 80.4% recall rate, with an absolute improvement of 1.2%.
