The geometry-based point cloud compression algorithm achieves efficient compression and transmission for LiDAR point clouds with high sparsity. However, the low-bitrate mode results in severe geometry compression artifacts, which involve both point reduction and coordinate offset. To the best of our knowledge, this is the first attempt to directly enhance the geometry quality for compressed LiDAR point cloud (CLGE) in a post-processing manner. Our proposed method consists of two branches: cylindrical densification and adaptive refinement. The former adopts a multi-scale sparse convolution framework to effectively extract spatial features in the cylindrical coordinate system and generate dense candidate points quickly. Large asymmetric sparse convolution kernels are also designed to capture the shapes of different regions and objects. The latter branch refines the candidate points through several MLP layers, which takes the neighborhood features between the candidate points and the input points into account. Finally, the designed ring-based farthest point resampling serves as an effective alternative for achieving the target number while maintaining the geometry distribution. Extensive experiments conducted on several datasets verify the effectiveness of our approach under different compression artifact levels. Furthermore, our method is easily extended to upsampling and is robust to noise. In addition to the geometry signal quality improvement, the point cloud enhanced by our proposed method alleviates the performance degradation in object detection task due to compression distortion.

Segmenting ultra-high-resolution (UHR) images poses a significant challenge due to constraints on GPU memory, leading to a trade-off between detailed local information and a comprehensive contextual understanding. Current UHR methods often employ a multi-branch encoder to handle local and contextual information, which can be memory-intensive. To address the need for both high accuracy and low memory usage in processing UHR images, we introduce a memory-efficient semantic segmentation approach by squeezing context information into local patches (SCPSeg). Our method integrates the processing of local and contextual information within a single-branch encoder. Specifically, we introduce a context squeezing module (CSM) designed to compress global context details into local patches, enabling segmentation networks to perceive broader image contexts. Additionally, we propose a super-resolution guided local feature alignment (LFA) technique to improve segmentation precision by aligning local feature relationships. This approach calculates similarities within sliding windows, avoiding heavy computational costs during the training phase. We evaluate the effectiveness of our proposed method on four widely used UHR segmentation benchmarks. Experimental results demonstrate that our approach enhances UHR segmentation accuracy without incurring additional memory overhead during the inference stage. The code is available at https://github.com/StuLiu/SCPSeg.

Cross-domain image segmentation plays a crucial role in the field of remote sensing. Current approaches often rely on a mean-teacher model that is integrated from student models to guide the training of the student model itself. However, the feature space of the mean-teacher model exhibits significant domain discrepancy and considerable class overlap, which results in suboptimal performance. Motivated by the idea of learning from stronger teachers, we introduce a robust domain adaptation method called LFMDA. This novel approach is the first to explicitly enhance cross-domain semantic segmentation performance by leveraging vision foundation models (VFMs) within remote sensing applications. Specifically, we propose a prototypical contrastive knowledge distillation loss (PCD) that enables the student model to produce domain-invariant yet category-discriminative features by distilling knowledge from a domain-generalized VFM teacher. Additionally, we introduce a local region homogenization strategy (LRH) to generate high-quality and high-quantity pseudo-labels by incorporating a Segment Anything Model (SAM). Extensive empirical evaluations demonstrate that our method outperforms existing approaches, setting a new state-of-the-art (SOTA) method in domain-adaptive remote sensing image segmentation. The code is available at https://github.com/StuLiu/LFMDA.

Real-time semantic segmentation plays an important role in auto vehicles. However, most real-time small object segmentation methods fail to obtain satisfactory performance on small objects, such as cars and sign symbols, since the large objects usually tend to devote more to the segmentation result. To solve this issue, we propose an efficient and effective architecture, termed small objects segmentation network (SOSNet), to improve the segmentation performance of small objects. The SOSNet works from two perspectives: methodology and data. Specifically, with the former, we propose a dual-branch hierarchical decoder (DBHD) which is viewed as a small-object sensitive segmentation head. The DBHD consists of a top segmentation head that predicts whether the pixels belong to a small object class and a bottom one that estimates the pixel class. In this situation, the latent correlation among small objects can be fully explored. With the latter, we propose a small object example mining (SOEM) algorithm for balancing examples between small objects and large objects automatically. The core idea of the proposed SOEM is that most of the hard examples on small-object classes are reserved for training while most of the easy examples on large-object classes are banned. Experiments on three commonly used datasets show that the proposed SOSNet architecture greatly improves the accuracy compared to the existing real-time semantic segmentation methods while keeping efficiency. The code will be available at https://github.com/StuLiu/SOSNet.

Recent years have witnessed the success of deep learning methods in quality enhancement of compressed point cloud. However, existing methods focus on geometry and attribute enhancement of single-frame point cloud. This paper proposes a novel compressed quality enhancement method for dynamic point cloud (DAE-MP). Specifically, we propose a fast inter-frame motion prediction module (IFMP) to explicitly estimate motion displacement and achieve inter-frame feature alignment. To maintain motion continuity between consecutive frames, we propose a motion consistency loss for supervised learning. Furthermore, a frequency component separation and fusion module is designed to extract rich frequency features adaptively. To the best of our knowledge, the proposed method is the first deep learning-based work to enhance the quality for compressed dynamic point cloud. Experimental results show that the proposed method can greatly improve the quality of compressed dynamic point cloud and provide a fast and efficient motion prediction plug-in for large-scale point cloud. For dynamic point cloud attribute with severely compressed artifact, our proposed DAE-MP method achieves up to 0.52dB (PSNR) performance gain. Moreover, the proposed IFMP module has a certain real-time processing ability for calculating the motion offset between dynamic point cloud frame.

Hyperspectral image (HSI) provides rich spatial and spectral information of ground objects, while synthetic aperture radar (SAR) records scattering information such as shape and structure. Fusion of HSI and SAR can improve the classification performance of land covers. In recent years, graph convolutional networks (GCN) have been widely used in the field of remote sensing due to its advantages in processing non-Euclidean structures, capturing local and global information. In this paper, we propose a spectral-spatial graph convolutional network (SSGCN) for fusion of HSI and SAR. First, the GCN is utilized to extract the spatial information of HSI and SAR. Then, a convolutional neural network is applied to extract the spectral information of HSI. Finally, the extracted spectral and spatial features are merged together followed by a fully connected layer to obtain the final classification result. Experiments on two datasets, i.e., Berlin and Augsburg, reveal that the proposed SSGCN significantly outperforms other representative methods.

Unsupervised domain adaption (UDA) has gained popularity in narrowing performance gaps across domains in remote sensing image semantic segmentation (RSISS). However, current UDA methods suffer from serious noisy pseudo-labels, adversely affecting domain adaptation performance. In this work, a local region homogenizing domain adaptation method (RegDA) is proposed to tackle this issue. Specifically, a generalized segment anything model (SAM) is utilized to obtain the semantic-consistent regions for the images in the target domain. Furthermore, a pixel-level voting scheme is proposed to get the semantic label for each local region and assign it to each pixel within this region. In this way, more reliable pseudo-labels are obtained and domain adaptation performance is improved. Experiment results on ISPRS datasets demonstrate that the proposed RegDA outperforms previous UDA approaches for RSISS. The code will be available at https://github.com/StuLiu/RegDA.

Unsupervised cross-scene hyperspectral image classification transfers the learnable knowledge from a labeled source scene to an unlabeled target scene. Currently, many statistical distribution alignment methods are introduced to mitigate domain discrepancy. However, these methods ignore the finer class specific structure which may cause negative transfer. To solve this issue, a prototype-based inter-intra domain alignment network is proposed for unsupervised cross-scene hyperspectral image classification. Specifically, a prototype-based inter-intra alignment method is proposed to narrow the feature distribution gap. Furthermore, an uncertainty estimation is developed to obtain highly reliable pseudo-labels in the target scene. Experiment results on several datasets imply that the proposed method outperform several cutting-edge unsupervised classification methods.

Making full use of spatial-temporal information is the key factor for removing compressed video artifacts. Recently, many deep learning-based compression artifact reduction methods have emerged. Among them, a series of methods based on deformable convolution have shown excellent capabilities in spatio-temporal feature extraction. However, local deformable offset prediction and pixel-wise inter-frame feature alignment in the unidirectional form limit the full utilization of temporal features in the existing method. Additionally, compressed video shows inconsistent degrees of distortion on different frequency components, and their restoration difficulty is also nonuniform. For the above problems presented by existing methods, we propose anenlarged motion-aware and frequency-aware network(EMAFA) to further extract spatio-temporal information and enhance information of different frequency components. To perceive different degrees of motion artifacts between compressed frames as accurately as possible, we design a bidirectional dense propagation pattern withpixel-wise and patch-wise deformable convolution(PIPA) module in the feature domain. In addition, we propose amulti-scale atrous deformable alignment(MSADA) module to enrich spatio-temporal features in image domain. Moreover, we design amulti-direction frequency enhancement(MDFE) module with multiple direction convolution to enhance the features of different frequency components. The experimental results show that the proposed method performs better than the state-of-the-art methods in both objective evaluation and visual perception experience. Supplementary experiments for Internet Streamed Video with hybrid-distortion demonstrate that our method also exhibits considerable generalizability for quality enhancement.

Oriented object detection with remote sensing scenes has made excellent progress in recent years, especially using anchor-free detectors. Without the limitation of inherent prior spatial information, anchor-free detectors regress the detection boxes from the object center or edge in an elegant way. However, anchor-free detectors suffer severe feature misalignment and inconsistency between classification and regression. Especially in remote sensing scenes, there are densely arranged instances and multi-scale representations, which will affect the detection accuracy. Therefore, a feature augmentation module (FAM) and an oriented feature alignment (OFA) module are proposed for oriented object detection called FAA-Det. More specifically, we first introduce a FAM to enhance the object representation. After that, the augmented feature maps will be fed into OFA for feature alignment and accurate detection. OFA has two independent branches for classification and regression, and their separate structures can alleviate the inconsistency in detection. FAM and OFA comprise the FAA-Head in our detector. Extensive evaluation demonstrates the effectiveness of our proposed FAA-Det that performs the state-of-the-art (SOTA) mean average precision (mAP) on the DOTA and HRSC2016 datasets without bells and whistles. Our code will be available athttps://github.com/jimuIee/FAA-Det.

Domain adaptive segmentation has recently gained more and more attention in the remote sensing field. However, current methods often generate a significant number of uncertain examples, i.e., noisy pseudo-labels, in the target domain, which adversely affects model convergence. To solve this issue, an uncertain example mining network is proposed for domain adaptive segmentation of remote sensing images. Specifically, a novel strategy called multilevel pseudo-label correcting (MPC) is proposed to correct the pseudo-labels in class, pixel, and superpixel levels. In this way, more reliable pseudo-labels can be selected for the subsequent training stage. Furthermore, a noise-robust example mining strategy, termed uncertainty-based valuable example mining (UVEM), is proposed to prioritize confident examples with significant gradients for training effectively. Extensive empirical evaluations on IsprsDA and LoveDA datasets demonstrate that the proposed method outperforms previous approaches, establishing state-of-the-art results in domain adaptive remote sensing image segmentation (RSIS). The code will be available athttps://github.com/StuLiu/UemDA.

In the past few years, there has been significant progress in hyperspectral image classification (HSIC). However, when the trained classifier on the source scene is directly applied to a new scene, the classification performance tends to dramatically decrease because of the spectral shift phenomenon. Most existing techniques use feature alignment to learn knowledge from labeled scenes to unlabeled scenes, often overlooking the impact of noisy samples and outliers. To tackle this issue, the classwise prototype-guided alignment network (CPGAN) is proposed for cross-scene HSIC. The core idea is that classwise prototypes across scenes are employed as alignment intermediaries to guide cross-scene feature alignment. Specifically, first, spectral-spatial features from different scenes are extracted with a common feature extractor. Then, an uncertainty-aware pseudolabel selection (UPS) is designed to obtain high-confidence pseudolabels for unlabeled target scenes. Finally, a novel classwise prototype-guided alignment method is proposed to simultaneously achieve interdomain and intradomain alignment (IntraDA). The experimental results conducted on three datasets show that our method achieves superior performance compared to other cutting-edge classification algorithms.

Hyperspectral image (HSI) classification methods have made great progress in recent years. However, most of these methods are rooted in the closed-set assumption that the class distribution in the training and testing stages is consistent, which cannot handle the unknown class in open-world scenes. In this work, we propose a feature consistency-based prototype network (FCPN) for open-set HSI classification, which is composed of three steps. First, a three-layer convolutional network is designed to extract the discriminative features, where a contrastive clustering module is introduced to enhance the discrimination. Then, the extracted features are used to construct a scalable prototype set. Finally, a prototype-guided open-set module (POSM) is proposed to identify the known samples and unknown samples. Extensive experiments reveal that our method achieves remarkable classification performance over other state-of-the-art classification techniques.