基于解耦RGB-D特征的六维物体位姿估计代码

我们提出了一种简单但有效的方法，名为MixedFusion。与以往的方法不同，我们认为颜色和点云的空间对应关系可以分开然后重新连接，从而实现更灵活的融合方案。通过执行所提出的方法，可以混合和融合更多信息丰富的颜色特征。我们在具有挑战性的LineMod和YCB-Video数据集上进行了大量实验，结果显示我们的方法显著提升了性能，而不引入额外的开销。此外，在度量的最小公差变窄时，所提出的方法对于高精度需求表现更佳。

We propose a simple yet effective method termed MixedFusion. Unlike existing approaches, we posit that the spatial correspondences between color data and point clouds can be decoupled and subsequently reconnected, thereby enabling a more flexible fusion framework. By applying the proposed method, more informative color features can be integrated and fused. We conduct extensive experiments on the challenging LineMod and YCB-Video datasets, and the results demonstrate that our method achieves significant performance improvements without incurring any additional computational overhead. Furthermore, when the minimum tolerance of the evaluation metric is narrowed, the proposed method performs even better for applications demanding high precision.