arXiv subjects

Jiro Katto

Publications and source records attributed to Jiro Katto.

At least 19 recordsLinked to original sources

arXiv2026

Streamable Neural Video Compression: A Mixed Precision Approach for Cross-Platform Deployment

Neural Video Codecs (NVCs) offer unprecedented rate-distortion performance, making them highly attractive for bandwidth-constrained environments like 5G cellular networks and emerging satellite direct-to-cell (D2C) links. However, deploying NVCs in real-world streaming applications is severely hindered by cross-platform floating-point non-determinism, which causes arithmetic entropy coders to desynchronize and crash across different GPU architectures. While recent integer-based quantization methods address this, they incur either massive degradation in compression efficiency (INT8) or severe computational bottlenecks by bypassing hardware acceleration (INT16). In this paper, we propose a streamable, client-server NVC architecture featuring a novel Mixed Precision (FP16/FP32) strategy. By strategically executing P-frames in hardware-accelerated FP16 for real-time throughput, while forcing I-frames and periodic feature-adapter resets to IEEE-754 compliant FP32, we guarantee deterministic synchronization at critical boundaries. Through extensive cross-encode/decode evaluations across 12 GPUs spanning four architectural generations, we demonstrate that our approach successfully eliminates intra-generation fragmentation and substantially broadens cross-die interoperability, achieving seamless cross-generation decodability for recent architectures at 1080p. Crucially, this is achieved with a negligible impact on compression efficiency. Furthermore, we evaluate the system's end-to-end latency across diverse real-world networks, including Wi-Fi 6, 5G NR (FDD/TDD), and Starlink D2C, proving the practical viability of streamable learned video compression while highlighting unique challenges in Non-Terrestrial Networks.

eess.IV↗

arXiv2026

Wavefront Parallelization for Efficient Learned Image Compression

Autoregressive context models are foundational for learned image compression,but they suffer from slow serial inference. Existing acceleration methods such as checkerboard context require architectural changes and retraining, thus are inapplicable to pre-trained models. We propose a completely training-free inference-time acceleration algorithm inspired by wavefront parallelism in video coding standards. Our method reorganizes inference into an optimal ``staggered'' wavefront order, minimizing sequential steps while maintaining exact autoregressive dependencies. Experimental results show our approach accelerates pre-trained autoregressive models (e.g., Cheng et al.) by more than $13\times$ while preserving the original rate-distortion performance. We also demonstrate that faster decoding is possible by trading off precise context dependencies. Source code will be available at https://github.com/tokkiwa/compressai-wavefront.

eess.IV↗

arXiv2026

Performance Analysis of Hardware-Accelerated 10-Bit 4:2:2 Encoding with Split-Frame Encoding for High-Fidelity V-PCC Streaming

Video-based Point Cloud Compression (V-PCC) encodes volumetric data by projecting 3D geometry and texture onto 2D video frames. To prevent spatial distortion and color bleeding during 3D reconstruction, this process requires 10-bit color depth and 4:2:2 chroma subsampling, rather than the standard 8-bit 4:2:0 format. Additionally, capturing high-density dynamic point clouds requires demanding encoding parameters, such as 8K resolution at framerates up to 120 fps. Historically, the lack of 4:2:2 chroma support in older GPU hardware encoders restricted real-time V-PCC to custom Application-Specific Integrated Circuits (ASICs). However, the recent introduction of NVIDIA's Blackwell GPU architecture, featuring on-chip hardware encoders with 10-bit 4:2:2 support, presents an opportunity to shift this workload to general-purpose hardware. This paper investigates the feasibility of such an approach. Using a commercially available Blackwell GPU equipped with four parallel on-die hardware encoders as a testbed, we evaluate the throughput, rate-distortion (RD) performance, and power consumption of 8K 10-bit 4:2:2 HEVC across various Split-Frame Encoding (SFE) configurations. Our results demonstrate that 4-way SFE achieves an encoding throughput of 122 fps, successfully meeting the strict real-time constraints of high-density V-PCC. Although the inability to exploit spatial redundancies across slice boundaries results in a BD-Rate penalty of up to 5%, the measured throughput and power efficiency establish standard, commercial off-the-shelf GPUs as a highly viable baseline for real-time volumetric video streaming.

eess.IV↗

arXiv2026

Dual-Constrained Diffusion Image Compression for Operational Rate-Distortion-Perception Optimization

The rate-distortion-perception (RDP) trade-off extends classical rate--distortion theory by imposing a distributional constraint on reconstructions, providing a unified framework for neural image compression that jointly governs fidelity and perceptual realism. While prior work achieves near-optimal rate--perception trade-offs, practical frameworks explicitly realizing the full RDP surface remain scarce, primarily due to the difficulty of introducing common randomness at the decoder. We propose DCIC (Dual-Constrained Diffusion Image Compression), which integrates a learned codec with a diffusion-based decoder governed by joint distortion and idempotence constraints. The distortion constraint bounds reconstruction fidelity relative to the base codec output; the idempotence constraint -- requiring that re-encoding the restored image recovers the base codec reconstruction -- serves as a tractable surrogate for the distributional perception requirement. Together, they steer the reverse denoising process via iterative optimization with consistent noise injection, realizing common randomness without additional rate overhead. At fixed rate, dual attenuation factors $(K_D, K_P)$ jointly navigate the Pareto frontier of the distortion-perception plane, enabling continuously adjustable fidelity-realism trade-offs from a single bitstream. DCIC$_{RD}$ ($K_P{=}0$) and DCIC$_{RP}$ ($K_D{=}0$) arise as boundary curves, with DCIC$_{RDP}$ ($K_D = K_P=1$) realizing the optimal interior operating point. Experiments on CelebA-HQ, CLIC2020, and ImageNet-1K across CNN, Transformer, and hybrid architectures confirm that DCIC$_{RDP}$ achieves superior BD-PSNR over all perceptual codecs, while DCIC$_{RP}$ matches dedicated perception-oriented methods in BD-FID, validating the practical value of full RDP surface navigation.

cs.CV↗

arXiv2026

Transformer-Based MCS Prediction for 5G Multicast-Broadcast Services (MBS)

The deployment of 5G Multicast-Broadcast Services (MBS) is emerging as a critical technology for spectral-efficient UHD content delivery and serving as a promising solution to modernize CATV deployment. However, unlike unicast networks that rely on RLC-AM with HARQ retransmissions, MBS broadcast operates in RLC Unacknowledged Mode (RLC-UM), where the absence of a feedback loop means packet loss is permanent and immediately impacts user QoE. Conventional link adaptation algorithms, designed for unicast, typically aggressively maximize throughput and fail in this risk-intolerant environment, resulting in severe video stalls and rebuffering. To address this, we propose a lightweight Transformer-based framework that predicts the success probability of all 28 MCS indices over an upcoming video segment horizon. Utilizing a unique commercial network dataset with 0.5 ms slot-level granularity, we train our model using a custom Asymmetric Safety Loss function that penalizes channel overestimation to prioritize link stability. Experimental results show that our approach achieves a reliability score of 86.89%, significantly outperforming standard AI baselines optimized for raw throughput (31.65%) while maintaining a safe conservative bias. Furthermore, the model is optimized for real-time applications, demonstrating an inference time of less than 0.07 ms on COTS 5G-era smartphones.

cs.NI↗

arXiv2026

Sustainable Real-Time 8K60 HEVC Encoding for V2X: Repurposing Legacy NVENC Hardware at the Vehicular Edge

The rapid advancement of Vehicle-to-Everything (V2X) communications and Tele-Operated Driving (ToD) demands ultra-low-latency, 8K60 video telemetry. However, deploying modern hardware at the vehicular edge is frequently hindered by supply chain constraints, high power budgets, and growing e-waste concerns. This paper investigates a highly sustainable alternative: repurposing legacy NVIDIA Pascal GPUs for real-time 8K HEVC edge encoding. We demonstrate that triggering 2-Way Split Frame Encoding (SFE) on dual-NVENC GP104 and GP102 silicon successfully unlocks real-time 8K60 throughput with a negligible Rate-Distortion penalty of under 1%. Crucially, our micro-architectural analysis reveals that smaller GPU dies significantly outperform larger flagship models in both raw throughput and energy efficiency. Because fixed-function encoding forces general-purpose Streaming Multiprocessor (SM) cores to sustain maximum frequencies while remaining idle, GPUs with fewer CUDA cores waste drastically less power. While benchmarking against the state-of-the-art RTX PRO 6000 Blackwell highlights a generational compression efficiency gap, Pascal's functional HEVC architecture and native lack of B-frames align perfectly with ultra-low-latency V2X pipelines. Ultimately, repurposed mid-range Pascal GPUs present a highly capable, cost-effective, and e-waste mitigating solution for modern Intelligent Transportation Systems.

eess.IV↗

arXiv2026

Evolution of NVENC Efficiency: A Longitudinal Analysis of HQ and UHQ Tuning Efficiency, Latency and Energy Trade-offs

The rapid expansion of uplink-intensive applications necessitates video coding solutions that balance high Rate-Distortion (RD) efficiency with ultra-low latency. This paper presents a longitudinal performance analysis of NVIDIA hardware encoding (NVENC), spanning from Pascal to the emerging Blackwell generation. We specifically evaluate the operational viability of the new "Ultra High Quality" (UHQ) tuning mode against standard low-latency configurations. Our results demonstrate that while the Blackwell architecture breaks historical efficiency plateaus, achieving a 5.94% BD-Rate gain in standard modes and up to 22.79% in UHQ modes, these gains incur severe system-level penalties. We reveal that UHQ operates as a hybrid pipeline, offloading complexity to CUDA cores and enforcing aggressive temporal structures (up to 7 B-frames) that increase end-to-end latency by over 400% and GPU board power consumption by up to 40%. Consequently, while UHQ successfully bridges the quality gap with software encoders, its prohibitive serialization delay renders it unsuitable for interactive real-time communications, positioning it instead as a specialized solution for Video-on-Demand (VoD) transcoding.

eess.IV↗

arXiv2025

Performance Evaluation of Low-Latency Live Streaming of MPEG-DASH UHD video over Commercial 5G NSA/SA Network

5G Standalone (SA) is the goal of the 5G evolution, which aims to provide higher throughput and lower latency than the existing LTE network. One of the main applications of 5G is the real-time distribution of Ultra High-Definition (UHD) content with a resolution of 4K or 8K. In Q2/2021, Advanced Info Service (AIS), the biggest operator in Thailand, launched 5G SA, providing both 5G SA/NSA service nationwide in addition to the existing LTE network. While many parts of the world are still in process of rolling out the first phase of 5G in Non-Standalone (NSA) mode, 5G SA in Thailand already covers more than 76% of the population. In this paper, UHD video will be a real-time live streaming via MPEG-DASH over different mobile network technologies with minimal buffer size to provide the lowest latency. Then, performance such as the number of dropped segments, MAC throughput, and latency are evaluated in various situations such as stationary, moving in the urban area, moving at high speed, and also an ideal condition with maximum SINR. It has been found that 5G SA can deliver more than 95% of the UHD video segment successfully within the required time window in all situations, while 5G NSA produced mixed results depending on the condition of the LTE network. The result also reveals that the LTE network failed to deliver more than 20% of the video segment within the deadline, which shows that 5G SA is absolutely necessary for low-latency UHD video streaming and 5G NSA may not be good enough for such task as it relies on the legacy control signal.

cs.NI↗

arXiv2025

Real-World Performance Evaluations of Low-Band 5G NR/4G LTE 4x4 MIMO on Commercial Smartphones

All 3GPP-compliant commercial 5G New Radio (NR)-capable UEs on the market are equipped with 4x4 MIMO support for Mid-Band frequencies (>1.7 GHz) and above, enabling up to rank 4 MIMO transmission. This doubles the theoretical throughput compared to rank 2 MIMO and also improves reception performance. However, 4x4 MIMO support on low-band frequencies (<1 GHz) is absent in every commercial UEs, with the exception of the Xperia 1 flagship smartphones manufactured by Sony Mobile and the Xiaomi 14 Pro as of January 2024. The reason most manufacturers omit 4x4 MIMO support for low-band frequencies is likely due to design challenges or relatively small performance gains in real-world usage due to the lack of 4T4R deployment on low-band by mobile network operators around the world. In Thailand, 4T4R deployment on the b28/n28 (APT) band is common on True-H and dtac networks, enabling 4x4 MIMO transmission on supported UEs. In this paper, the real-world 4x4 MIMO performance on the b28/n28 (APT) band will be investigated by evaluating the reliability test under different signal conditions and the maximum throughput test by evaluating the performance under optimal conditions, using the Sony Xperia 1 III and the Sony Xperia 1 IV smartphone. Devices from other manufacturers are also used in the experiment to investigate the performance with 2Rx antennas for comparison. Through firmware modifications, the Sony Xperia 1 III and IV can be configured to use only 2 Rx ports on low-band, enabling the collection of comparative 2 Rx performance data as a reference.

cs.NI↗

arXiv2025

Evaluations of High Power User Equipment (HPUE) in Urban Environment

While Time Division Duplexing (TDD) 5G New Radio (NR) networks offers higher downlink throughput due to the utilization of the middle frequency band, the uplink performance is negatively impacted due to higher path loss associated with higher frequencies, which degrade the users QoE in less optimal conditions. With the growing demand for high performance uplink throughput from novel applications such as Metaverse, Internet of Things (IoTs) and Smart City, 3GPP introduced High Power User Equipment (HPUE) on 5G TDD bands, allowing UEs to utilize more than 23 dBm of power for transmission to improve throughput, QoE, and reliability, especially at the cell edges. In this paper, the performance of HPUE is evaluated in the urban area on a commercial 5G network in terms of Uplink Throughput, Modulation Efficiency, Re-transmission Rate (ReTx Rate), and Power Consumption in both Standalone (SA) and Non-Standalone (NSA) modes. Through modem firmware modification, the performance is also compared across different power classes and antenna configurations.

cs.NI↗

arXiv2025

Performance Evaluation of Uplink 256QAM on Commercial 5G New Radio (NR) Networks

While Uplink 256QAM (UL-256QAM) has been introduced since 2016 as a part of 3GPP Release 14, the adoption was quite poor as many Radio Access Network (RAN) and User Equipment (UE) vendors didn't support this feature. With the introduction of 5G, the support of UL-256QAM has been greatly improved due to a big re-haul of RAN by Mobile Network Operators (MNOs). However, many RAN manufacturers charge MNOs for licenses to enable UL-256QAM per cell basis. This led to some MNOs hesitating to enable the feature on some of their gNodeB or cells to save cost. Since it's known that 256QAM modulation requires a very good channel condition to operate, but UE has a very limited transmission power budget. In this paper, 256QAM utilization, throughput and latency impact from enabling UL-256QAM will be evaluated on commercial 5G Standalone (SA) networks in two countries: Japan and Thailand on various frequency bands, mobility characteristics, and deployment schemes. By modifying the modem firmware, UL-256QAM can be turned off and compared to the conventional UL-64QAM. The results show that UL-256QAM utilization was less than 20% when deployed on a passive antenna network resulting in an average of 8.22% improvement in throughput. However, with Massive MIMO deployment, more than 50% utilization was possible on commercial networks. Furthermore, despite a small uplink throughput gain, enabling UL-256QAM can lower the latency when the link is fully loaded with an average improvement of 7.97 ms in TCP latency observed across various test cases with two TCP congestion control algorithms.

cs.NI↗

arXiv2025

Evaluation of NVENC Split-Frame Encoding (SFE) for UHD Video Transcoding

NVIDIA Encoder (NVENC) features in modern NVIDIA GPUs, offer significant advantages over software encoders by providing comparable Rate-Distortion (RD) performance while consuming considerably less power. The increasing capability of consumer devices to capture footage in Ultra High-Definition (UHD) at 4K and 8K resolutions necessitates high-performance video transcoders for internet-based delivery. To address this demand, NVIDIA introduced Split-Frame Encoding (SFE), a technique that leverages multiple on-die NVENC chips available in high-end GPUs. SFE splits a single UHD frame for parallel encoding across these physical encoders and subsequently stitches the results, which significantly improves encoding throughput. However, this approach is known to incur an RD performance penalty. The widespread adoption of NVIDIA GPUs in data centers, driven by the rise of Generative AI, means NVENC is poised to play a critical role in transcoding UHD video. To better understand the performance-efficiency tradeoff of SFE, this paper evaluates SFE's impact on RD performance, encoding throughput, power consumption, and end-to-end latency using standardized test sequences. The results show that for real-time applications, SFE nearly doubles encoding throughput with a negligible RD performance penalty, which enables the use of higher-quality presets for 4K and makes real-time 8K encoding feasible, effectively offsetting the minor RD penalty. Moreover, SFE adds no latency at 4K and can reduce it at 8K, positioning it as a key enabler for high-throughput, real-time UHD transcoding.

cs.AR↗

arXiv2025

Evaluation of GPU Video Encoder for Low-Latency Real-Time 4K UHD Encoding

The demand for high-quality, real-time video streaming has grown exponentially, with 4K Ultra High Definition (UHD) becoming the new standard for many applications such as live broadcasting, TV services, and interactive cloud gaming. This trend has driven the integration of dedicated hardware encoders into modern Graphics Processing Units (GPUs). Nowadays, these encoders support advanced codecs like HEVC and AV1 and feature specialized Low-Latency and Ultra Low-Latency tuning, targeting end-to-end latencies of < 2 seconds and < 500 ms, respectively. As the demand for such capabilities grows toward the 6G era, a clear understanding of their performance implications is essential. In this work, we evaluate the low-latency encoding modes on GPUs from NVIDIA, Intel, and AMD from both Rate-Distortion (RD) performance and latency perspectives. The results are then compared against both the normal-latency tuning of hardware encoders and leading software encoders. Results show hardware encoders achieve significantly lower E2E latency than software solutions with slightly better RD performance. While standard Low-Latency tuning yields a poor quality-latency trade-off, the Ultra Low-Latency mode reduces E2E latency to 83 ms (5 frames) without additional RD impact. Furthermore, hardware encoder latency is largely insensitive to quality presets, enabling high-quality, low-latency streams without compromise.

cs.AR↗

arXiv2025

Evaluation of Hardware-based Video Encoders on Modern GPUs for UHD Live-Streaming

Many GPUs have incorporated hardware-accelerated video encoders, which allow video encoding tasks to be offloaded from the main CPU and provide higher power efficiency. Over the years, many new video codecs such as H.265/HEVC, VP9, and AV1 were added to the latest GPU boards. Recently, the rise of live video content such as VTuber, game live-streaming, and live event broadcasts, drives the demand for high-efficiency hardware encoders in the GPUs to tackle these real-time video encoding tasks, especially at higher resolutions such as 4K/8K UHD. In this paper, RD performance, encoding speed, as well as power consumption of hardware encoders in several generations of NVIDIA, Intel GPUs as well as Qualcomm Snapdragon Mobile SoCs were evaluated and compared to the software counterparts, including the latest H.266/VVC codec, using several metrics including PSNR, SSIM, and machine-learning based VMAF. The results show that modern GPU hardware encoders can match the RD performance of software encoders in real-time encoding scenarios, and while encoding speed increased in newer hardware, there is mostly negligible RD performance improvement between hardware generations. Finally, the bitrate required for each hardware encoder to match YouTube transcoding quality was also calculated.

eess.IV↗

arXiv2025

FlashGMM: Fast Gaussian Mixture Entropy Model for Learned Image Compression

High-performance learned image compression codecs require flexible probability models to fit latent representations. Gaussian Mixture Models (GMMs) were proposed to satisfy this demand, but suffer from a significant runtime performance bottleneck due to the large Cumulative Distribution Function (CDF) tables that must be built for rANS coding. This paper introduces a fast coding algorithm that entirely eliminates this bottleneck. By leveraging the CDF's monotonic property, our decoder performs a dynamic binary search to find the correct symbol, eliminating the need for costly table construction and lookup. Aided by SIMD optimizations and numerical approximations, our approach accelerates the GMM entropy coding process by up to approximately 90x without compromising rate-distortion performance, significantly improving the practicality of GMM-based codecs. The implementation will be made publicly available at https://github.com/tokkiwa/FlashGMM.

eess.IV↗

arXiv2025

A Multi-Grid Implicit Neural Representation for Multi-View Videos

Multi-view videos are becoming widely used in different fields, but their high resolution and multi-camera shooting raise significant challenges for storage and transmission. In this paper, we propose MV-MGINR, a multi-grid implicit neural representation for multi-view videos. It combines a time-indexed grid, a view-indexed grid and an integrated time and view grid. The first two grids capture common representative contents across each view and time axis respectively, and the latter one captures local details under specific view and time. Then, a synthesis net is used to upsample the multi-grid latents and generate reconstructed frames. Additionally, a motion-aware loss is introduced to enhance the reconstruction quality of moving regions. The proposed framework effectively integrates the common and local features of multi-view videos, ultimately achieving high-quality reconstruction. Compared with MPEG immersive video test model TMIV, MV-MGINR achieves bitrate savings of 72.3% while maintaining the same PSNR.

eess.IV↗

arXiv2025

Real-time Video Prediction With Fast Video Interpolation Model and Prediction Training

Transmission latency significantly affects users' quality of experience in real-time interaction and actuation. As latency is principally inevitable, video prediction can be utilized to mitigate the latency and ultimately enable zero-latency transmission. However, most of the existing video prediction methods are computationally expensive and impractical for real-time applications. In this work, we therefore propose real-time video prediction towards the zero-latency interaction over networks, called IFRVP (Intermediate Feature Refinement Video Prediction). Firstly, we propose three training methods for video prediction that extend frame interpolation models, where we utilize a simple convolution-only frame interpolation network based on IFRNet. Secondly, we introduce ELAN-based residual blocks into the prediction models to improve both inference speed and accuracy. Our evaluations show that our proposed models perform efficiently and achieve the best trade-off between prediction accuracy and computational speed among the existing video prediction methods. A demonstration movie is also provided at http://bit.ly/IFRVPDemo. The code will be released at https://github.com/FykAikawa/IFRVP.

cs.CV↗

arXiv2025

Performance Analysis of 5G FR2 (mmWave) Downlink 256QAM on Commercial 5G Networks

The 5G New Radio (NR) standard introduces new frequency bands allocated in Frequency Range 2 (FR2) to support enhanced Mobile Broadband (eMBB) in congested environments and enables new use cases such as Ultra-Reliable Low Latency Communication (URLLC). The 3GPP introduced 256QAM support for FR2 frequency bands to further enhance downlink capacity. However, sustaining 256QAM on FR2 in practical environments is challenging due to strong path loss and susceptibility to distortion. While 256QAM can improve theoretical throughput by 33%, compared to 64QAM, and is widely adopted in FR1, its real-world impact when utilized in FR2 is questionable, given the significant path loss and distortions experienced in the FR2 range. Additionally, using higher modulation correlates to higher BLER, increased instability, and retransmission. Moreover, 256QAM also utilizes a different MCS table defining the modulation and code rate at different Channel Quality Indexes (CQI), affecting the UE's link adaptation behavior. This paper investigates the real-world performance of 256QAM utilization on FR2 bands in two countries, across three RAN manufacturers, and in both NSA (EN-DC) and SA (NR-DC) configurations, under various scenarios, including open-air plazas, city centers, footbridges, train station platforms, and stationary environments. The results show that 256QAM provides a reasonable throughput gain when stationary but marginal improvements when there is UE mobility while increasing the probability of NACK responses, increasing BLER, and the number of retransmissions. Finally, MATLAB simulations are run to validate the findings as well as explore the effect of the recently introduced 1024QAM on FR2.

cs.NI↗