arXiv subjects

Minnan Pei

Publications and source records attributed to Minnan Pei.

9 recordsLinked to original sources

arXiv2026

From Splats to Silicon: Rethinking Computational Efficiency of 3DGS

3D Gaussian splatting (3DGS) represents scenes with explicit primitives and supports real-time novel-view synthesis, yet its system efficiency varies substantially across scenes, viewpoints, rendering paths, and platform constraints. Existing studies pursue efficiency through representation and algorithm design, GPU runtime optimization, and architectural support, but their reported gains correspond to different points along the rendering and update paths. Connecting these indicators to end-to-end system benefit requires tracing how each optimization changes Gaussian selection, screen-space work, data movement, and stage or frame time. We therefore use a workload-centric framework to connect representation and algorithm research, GPU runtimes, and hardware architectures and to identify recurring workload patterns. We complement literature analysis with reproduced measurements and controlled GPU profiling of selected implementations, relating workload counts to stage time and memory traffic. Together, these comparisons show that system gains depend on workload reductions reaching downstream execution, granularity matching each stage, and the cost of data transfers, synchronization, and cached results, gradients, and optimizer data. Building on these findings, we discuss more consistent evaluation under rendering-quality constraints and identify key directions for future system design.

cs.AR↗

arXiv2026

DeGS: A Scalable 3DGS Architecture via Decoupled Workload Parsing and Reorganization

3D Gaussian Splatting (3DGS) has emerged as a leading technique for real-time novel view synthesis, yet existing 3DGS accelerators suffer from poor architectural scalability: increasing the number of PEs leads to marginal performance improvement during rendering. We identify that the root cause is the tightly coupled ``checking-while-blending'' dataflow, which exacerbates PE underutilization caused by spatial redundancy from irregular Gaussian coverage and temporal redundancy from asynchronous pixel-wise termination under parallel execution. To address this issue, we propose DeGS, a scalable architecture for efficient 3DGS inference. To systematically eliminate the redundancies inherent in rendering, DeGS exploits a decoupled dataflow, restructuring the coupled $\alpha$-checking, transmittance checking, and $\alpha$-blending of the standard rendering process into consecutive workload parsing, reorganization, and blending stages. This allows the fragmented, length-variable, and temporal-dependent workloads to be reorganized into compact, conflict-free, and dense workloads prior to blending, thereby significantly improving PE utilization during parallel blending. Implemented in 28 nm technology, DeGS achieves 2.36$\times$--7.25$\times$ throughput, 1.82$\times$--6.02$\times$ end-to-end speedup, and 1.59$\times$--4.42$\times$ energy efficiency over state-of-the-art 3DGS accelerators (GSCore, GBU, GCC) across diverse scenes and resolutions (720p to 8K). Moreover, scaling from 16 to 1024 PEs, DeGS maintains over 80\% PE utilization at high resolutions, significantly outperforming existing accelerators.

cs.AR↗

arXiv2026

A Motion-Aware Vector Quantization Framework with Centroid Reuse for Efficient VLA Inference

Vision-Language-Action (VLA) models have demonstrated strong potential for embodied AI, yet their high inference latency on GPUs limits real-time deployment. Existing accelerators, such as Dadu-Corki, improve efficiency but treat VLA models as full-precision workloads, leaving substantial redundancy in both memory and computation underexploited. In this paper, we propose VQVLA, an algorithm-hardware co-design framework that accelerates VLA inference by exploiting weight similarity and execution dynamics. We first introduce MotionVQ, a motion-aware vector quantization scheme that dynamically adjusts quantization precision based on the robot's execution state, reducing memory access while preserving task success rate. We then propose a merged-centroid vectorized GEMM paradigm that operates on the codebook-index representation, eliminating redundant multiplications through spatial aggregation and temporal reuse of centroids. To realize these optimizations, we design an accelerator that efficiently supports dynamic precision selection and centroid-reuse computation. Experimental results show that VQVLA achieves 6.5x, 2.8x, 1.9x, 3.3x, and 4.3x speedup over the A100 GPU, Dadu-Corki, LUT-DLA, CodeGEMM, and ShiftAddLLM, respectively, with negligible accuracy degradation.

cs.AI↗

arXiv2026

LATTICE: Constraint-Directed Scheduling, Memory Planning, and Pipeline Refinement for NPUs

General-purpose NPUs execute fine-grained command DAGs across heterogeneous compute and memory-transfer engines backed by finite, explicitly managed on-chip memories. This execution model creates a directed dependency between scheduling and memory planning: different legal topological orders induce different lifetime overlap, placement opportunities, and spill behavior, while a materialized layout introduces physical-address reuse constraints absent from the input precedence DAG. Command order therefore shapes the feasible memory plan, and the realized plan in turn defines the legal space for subsequent timing refinement. We present LATTICE, a deterministic constraint-directed compiler pipeline. Memory-Pressure-Aware Topological Scheduling reshapes lifetime geometry before address binding; Deterministic Linear Repackaging materializes tiered placement, spill/reload events, and plan-induced reuse constraints; and Critical Path Enhancement recovers pipeline parallelism while preserving the selected memory plan. Every accepted schedule passes independent memory and timing verification. Across six artifact-provided command traces labeled as derived from a Da Vinci NPU flow, LATTICE achieves the best or tied-best result in all 24 evaluated workload-metric comparisons. Relative to the best evaluated baseline for each workload and metric, it reduces peak memory, extra DDR traffic, spill count, and modeled makespan by 18.3% lower, 20.4% lower, 14.1% lower, and 16.3% lower, respectively. Plan-preserving CPE further reduces makespan by 12.1% over Freeze while leaving placement and memory traffic unchanged, establishing the static memory plan as a verifiable scheduling contract between memory planning and pipeline optimization.

cs.NI↗

arXiv2026

$\pi$-StepNFT: Wider Space Needs Finer Steps in Online RL for Flow-based VLAs

Flow-based vision-language-action (VLA) models excel in embodied control but suffer from intractable likelihoods during multi-step sampling, hindering online reinforcement learning. We propose \textbf{\textit{$\boldsymbol{\pi}$-StepNFT}} (Step-wise Negative-aware Fine-Tuning), a critic-and-likelihood-free framework that requires only a single forward pass per optimization step and eliminates auxiliary value networks. We identify that wider exploration spaces necessitate finer-grained, step-wise guidance for alignment. Empirically, $\pi$-StepNFT unlocks latent potential on LIBERO with competitive few-shot robustness. Moreover, it achieves superior generalization on ManiSkill, outperforming value-based baselines in OOD scenarios by preventing overfitting to multimodal features. This property offers a scalable solution promising for complex real-world applications.

cs.RO↗

arXiv2026

DALI: A Workload-Aware Offloading Framework for Efficient MoE Inference on Local PCs

Mixture of Experts (MoE) architectures significantly enhance the capacity of LLMs without proportional increases in computation, but at the cost of a vast parameter size. Offloading MoE expert parameters to host memory and leveraging both CPU and GPU computation has recently emerged as a promising direction to support such models on resourceconstrained local PC platforms. While promising, we notice that existing approaches mismatch the dynamic nature of expert workloads, which leads to three fundamental inefficiencies: (1) Static expert assignment causes severe CPUGPU load imbalance, underutilizing CPU and GPU resources; (2) Existing prefetching techniques fail to accurately predict high-workload experts, leading to costly inaccurate prefetches; (3) GPU cache policies neglect workload dynamics, resulting in poor hit rates and limited effectiveness. To address these challenges, we propose DALI, a workloaDAware offLoadIng framework for efficient MoE inference on local PCs. To fully utilize hardware resources, DALI first dynamically assigns experts to CPU or GPU by modeling assignment as a 0-1 integer optimization problem and solving it efficiently using a Greedy Assignment strategy at runtime. To improve prefetching accuracy, we develop a Residual-Based Prefetching method leveraging inter-layer residual information to accurately predict high-workload experts. Additionally, we introduce a Workload-Aware Cache Replacement policy that exploits temporal correlation in expert activations to improve GPU cache efficiency. By evaluating across various MoE models and settings, DALI achieves significant speedups in the both prefill and decoding phases over the state-of-the-art offloading frameworks.

cs.DC↗

arXiv2025

GCC: A 3DGS Inference Architecture with Gaussian-Wise and Cross-Stage Conditional Processing

3D Gaussian Splatting (3DGS) has emerged as a leading neural rendering technique for high-fidelity view synthesis, prompting the development of dedicated 3DGS accelerators for resource-constrained platforms. The conventional decoupled preprocessing-rendering dataflow in existing accelerators has two major limitations: 1) a significant portion of preprocessed Gaussians are not used in rendering, and 2) the same Gaussian gets repeatedly loaded across different tile renderings, resulting in substantial computational and data movement overhead. To address these issues, we propose GCC, a novel accelerator designed for fast and energy-efficient 3DGS inference. GCC introduces a novel dataflow featuring: 1) \textit{cross-stage conditional processing}, which interleaves preprocessing and rendering to dynamically skip unnecessary Gaussian preprocessing; and 2) \textit{Gaussian-wise rendering}, ensuring that all rendering operations for a given Gaussian are completed before moving to the next, thereby eliminating duplicated Gaussian loading. We also propose an alpha-based boundary identification method to derive compact and accurate Gaussian regions, thereby reducing rendering costs. We implement our GCC accelerator in 28nm technology. Extensive experiments demonstrate that GCC significantly outperforms the state-of-the-art 3DGS inference accelerator, GSCore, in both performance and energy efficiency.

cs.AR↗

arXiv2025

SpatialViz-Bench: A Cognitively-Grounded Benchmark for Diagnosing Spatial Visualization in MLLMs

Humans can imagine and manipulate visual images mentally, a capability known as spatial visualization. While many multi-modal benchmarks assess reasoning on visible visual information, the ability to infer unseen relationships through spatial visualization remains insufficiently evaluated as a spatial skill. This reliance on publicly sourced problems from IQ tests or math competitions risks data contamination and compromises assessment reliability. To this end, we introduce SpatialViz-Bench, a comprehensive multi-modal benchmark for spatial visualization with 12 tasks across 4 sub-abilities, comprising 1,180 programmatically generated problems, a scalable framework that allows for expansion to ensure fair and continuously reliable evaluations. Our evaluation of 27 Multi-modal Large Language Models (MLLMs) reveals wide performance variations, demonstrates the benchmark's strong discriminative power, and uncovers counter-intuitive findings: Chain-of-Thought (CoT) prompting paradoxically degrades accuracy on open-source models. Through statistical and qualitative analysis of error types, SpatialViz-Bench demonstrates that state-of-the-art MLLMs exhibit deficiencies in spatial visualization tasks, thereby addressing a significant lacuna in the field. The benchmark data and evaluation code are publicly available.

cs.CV↗

arXiv2022

Encoding feature supervised UNet++: Redesigning Supervision for liver and tumor segmentation

Liver tumor segmentation in CT images is a critical step in the diagnosis, surgical planning and postoperative evaluation of liver disease. An automatic liver and tumor segmentation method can greatly relieve physicians of the heavy workload of examining CT images and better improve the accuracy of diagnosis. In the last few decades, many modifications based on U-Net model have been proposed in the literature. However, there are relatively few improvements for the advanced UNet++ model. In our paper, we propose an encoding feature supervised UNet++(ES-UNet++) and apply it to the liver and tumor segmentation. ES-UNet++ consists of an encoding UNet++ and a segmentation UNet++. The well-trained encoding UNet++ can extract the encoding features of label map which are used to additionally supervise the segmentation UNet++. By adding supervision to the each encoder of segmentation UNet++, U-Nets of different depths that constitute UNet++ outperform the original version by average 5.7% in dice score and the overall dice score is thus improved by 2.1%. ES-UNet++ is evaluated with dataset LiTS, achieving 95.6% for liver segmentation and 67.4% for tumor segmentation in dice score. In this paper, we also concluded some valuable properties of ES-UNet++ by conducting comparative anaylsis between ES-UNet++ and UNet++:(1) encoding feature supervision can accelerate the convergence of the model.(2) encoding feature supervision enhances the effect of model pruning by achieving huge speedup while providing pruned models with fairly good performance.

eess.IV↗