A 3D point cloud completion task is to generate completed 3D objects given partial observations. Auto-encoder-based models suffer from poor generalization ability to untrained 3D data. Current diffusion-based models add isotropic noise with the same variance in three x, y, z axes. More importantly, these models ignore real-world anisotropic evolution properties of 3D particles from a non-equilibrium state to thermodynamic equilibrium in the real physical world due to the velocity and energy thermodynamics of the particles, leading to unstable completions of 3D object topology. This paper presents a novel physically-based anisotropic 3D diffusion model (3DDM) to address these issues. We also present derivations of our proposed forward and reverse processes and a loss function in closed form, thus reproducibility. The 3DDM contains anisotropic energy-aware forward and reverse processes with a novel anisotropic quadratic loss function. The forward process adds anisotropic 3D Gaussian noises per-axis and mimics the thermal non-equilibrium evolution towards Maxwellian equilibrium based on velocity and kinetic energy evolutions of 3D particles in the real physical space. The reverse process learns to denoise along per-axis and per-timestep anisotropically. The anisotropic quadratic loss function penalizes errors along certain axes, yielding a highly flexible and anisotropic reverse diffusion process and a physically realistic generative model. The 3DDM denoises along x, y, z axes with different velocities from the non-equilibrium evolution, achieving fewer than 20 diffusion steps and strong generalization to unseen 3D objects and real-world scenes that were not trained.

A 3D point cloud completion task is to generate one or multiple completed 3D objects based on incomplete and partial input data. While single-output models suffer from the large missing input data, current multiple-output models require texts, images or depth maps to guide completion tasks, dramatically increasing network model sizes with limited output diversities. This paper proposes a novel probabilistic architecture, MCNet, to address these challenges. MCNet includes a latent shape prior module and a conditional probabilistic diffusion module to explore multiple completion results faithful to the partial input data. We hypothesise that the learnt latent space contains diverse latent shapes. Therefore, through a reverse process of latent shape prior distribution, we can recover a diverse range of latent shapes from the random samples of a Gaussian distribution. Based on the diverse latent shapes and the latent shape of the partial input, we design a conditional probabilistic diffusion model to convert a noise distribution into the distribution of multiple 3D shapes. Critically, MCNet is lightweight and generates flexible output point cloud resolutions without any further training while maintaining the model size constant with the increase of the input and output point cloud resolutions. Experimental results demonstrate that MCNet achieves state-of-the-art performance and also exhibits strong generalization to unseen 3D objects and real-world 3D scenes that are never trained. Code is available at https://github.com/LONG-XI/MCNet .

Collective Location Selection (CLS) aims to identify$k$optimal sites for facility establishment to collectively maximize user attraction. Traditional CLS approaches often overlook user mobility and inter-facility competition, critical factors in real-world scenarios. This paper introduces MC2LS, the first effort on CLS that addresses these gaps by considering user mobility and peer competition. Solving MC2LS is nontrivial due to its NP-hardness. To overcome the challenge of pruning multi-point users with highly overlapping minimum boundary rectangles (MBRs), we develop a position count threshold and two square-based pruning rules. We propose IQuad-tree, a user-MBR-free index, to benefit the hierarchical and batch-wise properties of the pruning rules. We present an$(1-\frac{1}{e})$-approximate greedy solution to MC2LS, and empirical studies demonstrate the superiority of our proposed solution over the state-of-the-art techniques.

Collective Location Selection (CLS) has received significant research attention in the spatial database community due to its wide range of applications. The CLS problem selects a group ofkpreferred locations among candidate sites to establish facilities, aimed at collectively attracting the maximum number of users. Existing studies commonly assume every user is located in a fixed position, without considering the competition between peer facilities. Unfortunately, in real markets, users are mobile and choose to patronize from a host of competitors, making traditional techniques unavailable. To this end, this paper presents the first effort on a CLS problem in competition scenarios, calledmc$^{2}$2ls, taking into account the mobility factor. Solvingmc$^{2}$2lsis a non-trivial task due to its NP-hardness. To overcome the challenge of pruning multi-point users with highly overlapped minimum boundary rectangles (MBRs), we exploit a position count threshold and design two square-based pruning rules. We introduce IQuad-tree, a user-MBR-free index, to benefit the hierarchical and batch-wise properties of the pruning rules. We propose an$(1-\frac{1}{e})$-approximate greedy solution tomc$^{2}$2lsand incorporate a candidate-pruning strategy to further accelerate the computation for handling skewed datasets. Extensive experiments are conducted on real datasets, demonstrating the superiority of our proposed pruning rules and solution compared to the state-of-the-art techniques.

In the process of coal mine production, coal mine underground video surveillance is of great significance for enhancing coal mine safety production by analyzing and predicting sudden abnormalities in the coal mine production process in advance. However, it is difficult to ensure the safety of personnel by relying solely on human labor, and along with a large number of deep learning network models being proposed in recent years, it has become a challenge to apply target detection to actual mineral production environments. In this paper, we propose an improved RTMDet-Based target detection algorithm for underground coal mine to solve the problems of lack of real-time performance of video surveillance and restricted edge devices that are not easy to be deployed by combining knowledge distillation with target detection model. In order to verify the effectiveness of the method, the improved RTM-CMD model is compared with the RTMDet model and several mainstream target detection models in a comparison experiment. The experimental results show that the improved RTM-CMD model does not cause any significant loss of accuracy compared to the benchmark model in a resource-constrained environment, while still ensuring high accuracy compared to the rest of the models.