Point cloud segmentation forms the foundation of 3D scene understanding. Boundaries, the intersections of regions, are prone to mis-segmentation. Current point cloud segmentation models exhibit unsatisfactory performance on boundaries. There is limited focus on explicitly addressing semantic segmentation of point cloud boundaries. We introduce a method called Multi-fineness Boundary Constraint (MBC) to tackle this challenge. By querying boundaries at various degrees of fineness and imposing feature constraints within these boundary areas, we enhance the discrimination between boundaries and non-boundaries, improving point cloud boundary segmentation. However, solely emphasizing boundaries may compromise the segmentation accuracy in broader non-boundary regions. To mitigate this, we introduce a new concept of point cloud space termed ensemble and a Shifted Ensemble-aware Perception (SEP) module. This module establishes information interactions between points with minimal computational cost, effectively capturing direct point-to-point long-range correlations within ensembles. It enhances segmentation performance for both boundaries and non-boundaries.

Few-shot learning is widely applied in the current stage for remote sensing image classification to use prior knowledge to identify new classes faster. However, since existing few-shot remote sensing image classification methods only process the feature vectors extracted from the complete image, ignoring the localized knowledge in the input incorporated into the target can better utilize the contextual information to capture more local details. To address these problems, we propose a metric learning framework based on maximizing mutual information between different views (MMI-ML). Specifically, we introduce a self-supervised model to train an embedding network with enhanced feature representation by maximizing mutual information of global features and local features at different scales. In addition, we design a new embedding network to make it more appropriate for the self-supervised model. Finally, we devise a new loss function in the training stage, which can effectively speed up the convergence of the model. We conduct comparative experiments on three public remote sensing datasets, and the experimental results show that the classification accuracy of the MMI-ML framework is improved by up to 3.22%.

Existing methods of cross-modal domain adaptation for 3D semantic segmentation predict results only via 2D-3D complementarity that is obtained by cross-modal feature matching. However, as lacking supervision in the target domain, the complementarity is not always reliable. The results are not ideal when the domain gap is large. To solve the problem of lacking supervision, we introduce masked modeling into this task and propose a method Mx2M, which utilizes masked cross-modality modeling to reduce the large domain gap. Our Mx2M contains two components. One is the core solution, cross-modal removal and prediction (xMRP), which makes the Mx2M adapt to various scenarios and provides cross-modal self-supervision. The other is a new way of cross-modal feature matching, the dynamic cross-modal filter (DxMF) that ensures the whole method dynamically uses more suitable 2D-3D complementarity. Evaluation of the Mx2M on three DA scenarios, including Day/Night, USA/Singapore, and A2D2/SemanticKITTI, brings large improvements over previous methods on many metrics.

Weak target recognition, tracking and track management with a low signal-to-noise ratio (SNR) are always tricky problems. Probability hypothesis density (PHD) filtering propagates the first-order multi-target moment to obtain the best Poisson approximation to multi-target density. The PHD filtering does not consider explicit associations between measurements and targets, which is computationally efficient. But it cannot distinguish different targets or extract the time series of track states. Based on track-before-detect (TBD) strategies, this paper proposes labeled PHD (LPHD) filtering and derives its close-form solution, which identifies targets with a unique label. It is derived based on rigorous Bayes criteria, finite set statistics and Kullback-Leibler divergence minimization approximation. The separable TBD-based observation likelihood is conjugate to the Poisson mixture prior for LPHD filtering. Under the point-target assumption, the multi-hypothesis assignments of pixel-to-target are implemented with Murty’s K-shortest path algorithm for LPHD filtering. Additionally, sequential Monte Carlo (SMC) implementations under the nonlinear non-Gaussian assumption are devised. Finally, simulations exhibit good performance in low SNR scenarios.

Mainstream real-time instance segmentation methods always predict masks in the ’detect-then-segment’ way and ignore the object boundaries, leading to resource wasting and indistinct masks. To overcome these drawbacks, we propose a Dynamic Object Boundary-refinement Network (DOBNet) to predict masks in the principle of SOLO [1]. In this method, we first adapt the OctConv [2] as the generator to produce two parallel dynamic convolutions for mask and boundary features, respectively. The Boundary Refinement Module then helps fuse the features from the two convolutions and thereby refine the final predictions with boundary information. Hence, our method attains a precise segmentation while maintaining real-time speed. More specifically, the architecture achieved 37.9 AP on the COCO test-dev2017 dataset with a speed of 31.8 FPS, as is shown in Table 1. The results are more accurate than the existing real-time method.

Recently, in order to deploy neural networks on mobile devices, many studies have focused on reducing the number of parameters and computational complexity of neural networks. However, most existing methods do reduce the computational complexity of deep neural networks, but also greatly sacrifice their performance. To maintain the relative balance of computational complexity and performance of deep neural networks, this paper proposes a Global Attention Augmentation Ghost(GAAG) module, which decreases the number of parameters while bring performance improvements. Analyzed the network architecture of the Ghost module, we empirically show it is waste of computational resources that cheap operation in Ghost module produces more feature maps by linear transformations, only increasing the width of convolution neural network instead of extracting more useful feature information, and in a convolution layer composed of multiple feature blocks, the circulation of channel information is essential to better integrate the information of each feature blocks. Therefore, we propose a lightweight long-range dependency extraction block instead of cheap operation, which increases the ability to extract the non-local information of the model while keeping the computational cost almost invariable. Furthermore, we combine channel shuffle and channel attention to promote the fusion of local and non-local information. The proposed GAAG module is efficient yet effective and can be flexibly plugged into existing convolutional neural networks. Experiments conducted on the benchmark demonstrate that the GAAG module can perfectly replace the traditional convolutional layer in the baseline model. We extensively evaluate our GAAG module on image classification and object detection with backbones of ResNets. The experimental results show that our GAAG module can keep a good balance between lightweight and high performance compared with the similar model.