Referring image segmentation (RIS) aims to segment objects in an image conditioning on free-form text descriptions. Despite the overwhelming progress, it still remains challenging for current approaches to perform well on cases with various text expressions or with unseen visual entities, limiting its further application. In this paper, we present a novel RIS approach, which substantially improves the generalization ability by addressing the two dilemmas mentioned above. Specially, to deal with unconstrained texts, we propose to boost a given expression with an explicit and crucial prompt, which complements the expression in a unified context, facilitating target capturing in the presence of linguistic style changes. Furthermore, we introduce a multi-modal fusion aggregation module with visual guidance from a powerful pretrained model to leverage spatial relations and pixel coherences to handle the incomplete target masks and false positive irregular clumps which often appear on unseen visual entities. Extensive experiments are conducted in the zero-shot cross-dataset settings and the proposed approach achieves consistent gains compared to the state-of-the-art, e.g., $4.15 \%$, $5.45 \%$, and $4.64 \%$ mIoU increase on RefCOCO, RefCOCO+ and ReferIt respectively, demonstrating its effectiveness.

Consistent correspondences between point clouds are vital to 3D vision tasks such as registration and recognition. In this paper, we present a mutual voting method for ranking 3D correspondences. The key insight is to achieve reliable scoring results for correspondences by refining both voters and candidates in a mutual voting scheme. First, a graph is constructed for the initial correspondence set with the pairwise compatibility constraint. Second, nodal clustering coefficients are introduced to preliminarily remove a portion of outliers and speed up the following voting process. Third, we model nodes and edges in the graph as candidates and voters, respectively. Mutual voting is then performed in the graph to score correspondences. Finally, the correspondences are ranked based on the voting scores and top-ranked ones are identified as inliers. Feature matching, 3D point cloud registration, and 3D object recognition experiments on various datasets with different nuisances and modalities verify that MV is robust to heavy outliers under different challenging settings, and can significantly boost 3D point cloud registration and 3D object recognition performance.

Recent learning-based models excel in point cloud registration for low-overlap scenes but falter in scenarios with minimal overlap. In this article, we propose a novel method to address the extreme case of low-overlap registration: non-overlapping point cloud registration. This scenario involves input point clouds that do not have overlapping regions but are adjacent to each other after registration. While the practical application value of non-overlapping point cloud registration remains to be explored, we believe that researching this issue contributes to enhancing the performance of registration in scenarios with extremely low overlap. Abandoning conventional overlapping region detection, we directly generate the registered source point cloud with SCREAM, a generative adversarial network (GAN). The generator incorporates information from the target point cloud into the source point cloud’s features and generates the registered source point cloud. To further align the generated results with the target point cloud, we propose a differentiable renderer that renders both the target and predicted point clouds into depth maps. These depth maps are then used as inputs to a discriminator to determine whether the generated results align with the target point cloud. Rigid transformation can be directly estimated from the correspondences between the source and the generated point clouds, bypassing the need for detecting overlapping regions, feature matching, and RANSAC steps found in previous methods. Extensive experiments demonstrate that SCREAM not only outperforms common overlapping point cloud registration scenarios but also achieves a registration success rate of 52.6% for the first time in non-overlapping scenes. We also constructed a new indoor scene registration dataset, 3DZeroMatch, specifically designed to explore non-overlapping registration problems. Our code and the dataset 3DZeroMatch are accessible athttps://github.com/xujiabo/SCREAM/.