The paper presents memory‑efficient GPU pipelines that accelerate real‑time non‑line‑of‑sight (NLOS) reconstruction. By redesigning two wave‑based algorithms—f‑k migration and phasor‑fields—with fused kernels, warp‑level photon binning, batched transforms, CUDA graph replay, and selective FP16 storage, the authors achieve up to 42× speed‑ups over a reference streaming pipeline and 14× over the fastest published GPU baseline while reducing memory usage to as little as 2.5%. The work also includes an ablation study of implementation choices and introduces three denoising strategies that leverage the increased frame budget for future NLOS video processing.
arXiv:2506. 07069v2 Announce Type: replace-cross Abstract: 3D Gaussian Splatting (3DGS) has emerged as a powerful technique for novel view synthesis, combining high-quality reconstruction with efficient rendering.
arXiv:2608. 19639v1 Announce Type: new Abstract: Streaming reconstruction of Free-Viewpoint Videos (FVVs) supports immersive Internet of Things (IoT) services, such as telepresence and digital twin visualization.
Streaming reconstruction of Free-Viewpoint Videos (FVVs) supports immersive Internet of Things (IoT) services, such as telepresence and digital twin visualization. Existing methods suffer from high per-frame optimization time and large storage footprints, limiting deployment on resource-constrained Edge-IoT devices.
ReCoSplat is an online feed‑forward Gaussian splatting model that can synthesize novel views from a stream of observations, handling both posed and unposed inputs and optionally using camera intrinsics. It introduces a Render‑and‑Compare module that renders the current scene from the viewpoint of the incoming observation and compares it to the observation, providing a stable conditioning signal to mitigate the mismatch caused by predicted camera poses. A hybrid KV‑cache compression strategy further reduces memory usage, enabling the model to process long sequences efficiently while achieving state‑of‑the‑art performance on online view synthesis tasks.
arXiv:2608. 11271v1 Announce Type: cross Abstract: Next-generation Synthetic Aperture Radar (SAR) missions will generate data far faster than they can downlink, making onboard data reduction essential for near-real-time Earth observation.
ABCD (Alpha‑Composited Block Coordinate Descent) is an out‑of‑core training framework for alpha‑composited radiance fields, demonstrated on 3D Gaussian Splatting. It reformulates training as block coordinate descent over spatial partitions, keeping only one block of parameters active while pre‑rendering and collapsing inactive regions into foreground and background RGBA images. This approach reduces peak VRAM usage to a constant with respect to scene size, enabling training on GPUs with limited memory while maintaining reconstruction quality within 5% PSNR of 3DGS.
arXiv:2607. 23264v1 Announce Type: cross Abstract: Fine-grained, device-initiated communication lets persistent GPU kernels in distributed diffusion transformer (DiT) inference issue remote stores and overlap data movement with Tensor Core computation.
Real-time video restoration (VR) for live streams requires high-resolution outputs under strict per-frame latency constraints. Existing one-step diffusion-based VR models remain difficult to deploy on consumer-grade GPUs due to two main bottlenecks: quadratic spatial attention at high resolutions and the latency-memory overhead of large video autoencoders.
The paper investigates how a single SM utilization metric can misrepresent the true workload of large language model (LLM) inference on Nvidia Hopper GPUs. By profiling vLLM with FlashAttention‑3 and cuBLASLt on an H100 NVL across various phases (cold prefill, warm prefill, and decode) and varying sequence length and batch size, the authors replace the single utilization figure with eight detailed counter‑validated views. These views, tied to specific Nsight Compute counters or formulas, reveal how factors such as fragment fill, occupancy limits, stall signatures, wave quantization, and kernel selection create utilization gaps across four production models and six per‑layer kernel roles.
arXiv:2607. 14898v1 Announce Type: cross Abstract: Real-time video generation demands fast decoding as much as fast denoising, yet current latent video diffusion models rely on 3D convolutional decoders that are slow and memory-intensive at high resolutions or for long video.
arXiv:2606. 31938v1 Announce Type: cross Abstract: Deploying Vision Transformer (ViT) models on edge platforms remains challenging due to their high computational demands and the architectural heterogeneity of modern hybrid ViT models, which incorporate both fully connected and convolutional layers.
arXiv:2606. 11390v1 Announce Type: cross Abstract: Gaussian splatting methods have become increasingly popular for neural reconstruction of the real world.