Point clouds are a fundamental format for immersive experiences, posing significant challenges for storage and transmission. Unlike 2D image compression, 3D point clouds are sparse and irregular, complicating their attribute compression. While sparse convolution-based methods have made significant success on point cloud attribute compression by leveraging the sparsity of point clouds, they are constrained by a limited receptive field and insufficient adaptability to diverse inputs. To overcome these limitations, this paper proposes a novel point transformer-based architecture to exploit correlations and aggregate features across multiple scales (CSFormer). It retains the advantage of sparse convolution operating on occupied voxels and leverages varied sparsity distributions and the geometry distortions inherent in consecutive scales to construct attention maps, effectively extending the receptive field and adapting to different inputs. We further introduced GCPEM, a Geometry-aware Context Prediction-based Entropy Model that reduces bitrates by jointly utilizing the spatial and channel dependencies. Unlike previous methods that capture only one type of the information, GCPEM organizes latent features into groups interlaced across both space and channel dimensions and employs a context-prediction mechanism guided by known geometry for efficient coding. Experimental results show that the proposed method outperforms the state-of-the-art learning-based method and MPEG standard G-PCC codec over 7% and 28% in BD-BR (Y-PSNR), respectively. It has a time complexity comparable to the state-of-the-art learning-based method and the G-PCC. The source code and trained models will be released at https://github.com/X-H-offical/CST-PCAC-plus.git.

Point clouds serve as a fundamental format for virtual and augmented reality applications. However, their substantial data volumes present considerable challenges regarding data storage and transmission. This paper introduces a novel point transformer-based approach for point cloud attribute compression, termed as CST-PCAC, which utilizes a transformer architecture designed to exploit correlations across multiple scales. Our method is grounded in a variational autoencoder framework, enhanced by stacked sparse convolution (SConv) and the proposed Cross-Scale Point Transformer (CSFormer). CSFormer incorporates two types of attention mechanisms: Within-Scale Attention (WSA) and Cross-Scale Attention (CSA). WSA constructs attention maps within K-nearest neighbours at the current scale, while CSA operates across adjacent upper or lower scales. By leveraging varied sparsity distributions and the geometry distortions inherent in consecutive scales, CSFormer enhances the receptive field and feature representation, thereby effectively extracting correlations to minimize redundancy. Experimental results demonstrate that CST-PCAC achieves average Bjøntegaard Delta bitrate gains exceeding 30%, 10%, and 20% in comparison to Sparse-PCAC, Scalable-PCAC, and G-PCC, respectively. In terms of computational efficiency, CST-PCAC exhibits a time complexity two orders of magnitude lower than that of deep learning-based methods including Sparse-PCAC and Scalable-PCAC, while maintaining a comparable complexity to traditional methods like G-PCC.

The evolution of 3D visualization techniques has fundamentally transformed how we interact with digital content. At the forefront of this change is point cloud technology, offering an immersive experience that surpasses traditional 2D representations. However, the massive data size of point clouds presents significant challenges in data compression. Current methods for lossy point cloud attribute compression (PCAC) generally focus on reconstructing the original point clouds with minimal error. However, for point cloud visualization scenarios, the reconstructed point clouds with distortion still need to undergo a complex rendering process, which affects the final user-perceived quality. In this paper, we propose an end-to-end deep learning framework that seamlessly integrates PCAC with differentiable rendering, denoted as rendering-oriented PCAC (RO-PCAC), directly targeting the quality of rendered multiview images for viewing. In a differentiable manner, the impact of the rendering process on the reconstructed point clouds is taken into account. Moreover, we characterize point clouds as sparse tensors and propose a sparse tensor-based transformer, called SP-Trans. By aligning with the local density of the point cloud and utilizing an enhanced local attention mechanism, SP-Trans captures the intricate relationships within the point cloud, further improving feature analysis and synthesis within the framework. Extensive experiments demonstrate that the proposed RO-PCAC achieves state-of-the-art compression performance, compared to existing reconstruction-oriented methods, including traditional, learning-based, and hybrid methods. The code will be released athttps://github.com/net-F/RO-PCAC.git.

Digital Computing-in-Memory (DCIM) is an emerging architecture that integrates digital logic into memory for efficient AI computing. However, current DCIM designs heavily rely on manual efforts. This increases DCIM design time and limits the optimization space, making it challenging to satisfy the user specifications of diverse AI applications. This paper presents AutoDCIM, the first automated DCIM compiler. Au-toDCIM takes the user specifications as inputs and generates a DCIM macro architecture with an optimized layout. AutoDCIM’s template-based generation balances handcrafted cell design and agile macro development. AutoDCIM’s layout exploration loop analyzes diverse DCIM array partitioning schemes to satisfy user specifications. The auto-generated DCIM macros present competitive efficiency results in comparison with state-of-the-art silicon-verified DCIM macros.

Deep neural networks (DNNs) usually have multiple layers and thousands of trainable parameters to ensure high accuracy. Due to the requirement of large amounts of computation and memory, these networks are not suitable for real-time and resource-constrained mobile or embedded systems. Various techniques such as network pruning, weight sharing, network quantization, and weight encoding have been proposed to improve computational and memory efficiency. This paper presents a synchronous weight quantization-compression (SWQC) technique to compress the weights of low-bit quantized neural network (QNN). Specifically, it quantizes the weights not strictly according to their values but based on compression efficiency and their probabilities of being different quantized results. In the process of weight quantization, the compression efficiency of weights is considered as an important factor. With the help of retraining, a high compression rate and accuracy can be achieved. Verification is performed on 4-bit QNNs using the MNIST and CIFAR10 datasets. Results show that no classification accuracy is lost when the compression rate approaches 5.4X and 4.4X for the two datasets, respectively. The compression rate of the MNIST experiment is increased to 12.1X with a 1% accuracy drop, while the CIFAR10 experiment achieves a compression rate of 5.6X with the accuracy drop of about 0.6%.

MOTIVATION: Drop-seq has recently emerged as a powerful technology to analyze gene expression from thousands of individual cells simultaneously. Currently, Drop-seq technology requires refinement and quality control (QC) steps are critical for such data analysis. There is a strong need for a convenient and comprehensive approach to obtain dedicated QC and to determine the relationships between cells for ultra-high-dimensional datasets. RESULTS: We developed Dr.seq, a QC and analysis pipeline for Drop-seq data. By applying this pipeline, Dr.seq provides four groups of QC measurements for given Drop-seq data, including reads level, bulk-cell level, individual-cell level and cell-clustering level QC. We assessed Dr.seq on simulated and published Drop-seq data. Both assessments exhibit reliable results. Overall, Dr.seq is a comprehensive QC and analysis pipeline designed for Drop-seq data that is easily extended to other droplet-based data types. AVAILABILITY AND IMPLEMENTATION: Dr.seq is freely available at: http://www.tongji.edu.cn/∼zhanglab/drseq and https://bitbucket.org/tarela/drseq CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.