Rate-distortion optimization for full-reference image quality metrics via stochastic Hessian estimates
Summary
This paper proposes a method to integrate full-reference image quality metrics into rate-distortion optimization for video codecs by approximating them with input-dependent quadratic distortions using stochastic Hessian estimates, achieving BD-rate savings in VVC.
View Cached Full Text
Cached at: 09/25/26, 03:46 AM
Paper page - Rate-distortion optimization for full-reference image quality metrics via stochastic Hessian estimates
Source: https://huggingface.co/papers/2609.30077
Abstract
Block-basedvideocodecsselectcodingparametersbasedontheinputbyoptimizingarate-distortiontrade-off.Theconventionaldistortionchoice,thesumofsquarederrors(SSE),simplifiesparameterselection:theSSEisthesumofblock-wiseSSEs,sorate-distortionoptimization(RDO)cantreatblocksindependently.Alternatively,full-referenceimagequalityassessment(FR-IQA)metricssuchasMS-SSIMorLPIPSoftenalignbetterwiththehumanvisualsystemthanSSE,buttheycannotbeusedin-loop:theydonotdecomposeblock-wiseandtypicallyrequirethefullydecodedimageasinput.Buildingonexistingresultsinmetricquadratization,weapproximateabroadclassofFR-IQAmetricsbyaninput-dependentquadraticdistortion(IDQD),whosequadraticformmatrixisderivedfromtheHessianofthemetricevaluatedatthesourcevideo.Tomakethedistortioncomputableblock-wise,weproposetwoapproximationsoftheHessianmatrix:1)keepingtheblock-diagonal,and2)keepingonlyitsdiagonal.Weproposeestimatorsforboththatrequireonlymatrix-vectorproductswiththeHessianobtainedbyautomaticdifferentiation.AcrossfivemetricsforKodakandCLICinVVC,IDQD-RDOachieves14.2-36.7%BD-ratesavingsunderthetargetmetricwithnodecoderchangesandincurs10-30%encodingcomplexityoverhead.
View arXiv pageView PDFGitHub0Add to collection
Get this paper in your agent:
hf papers read 2609\.30077
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2609.30077 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2609.30077 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2609.30077 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
RateQuant: Optimal Mixed-Precision KV Cache Quantization via Rate-Distortion Theory
This paper introduces RateQuant, a method for optimal mixed-precision KV cache quantization that uses rate-distortion theory to address distortion model mismatch. It significantly reduces perplexity compared to existing methods like KIVI and QuaRot with minimal calibration overhead.
A Unified Rate-Distortion Perspective on Vector, Product, and Scalar Quantization
This paper proposes a unified rate-distortion perspective on discrete visual tokenization, resolving key questions about quantization objectives and comparisons, and shows that vector quantization achieves the lowest distortion under controlled conditions.
Quantized Keys Steal Attention: Bias Correction for KV-Cache Compression in Video Diffusion
This paper identifies a bias in attention weights caused by quantizing keys in KV-cache compression for chunk-wise autoregressive video diffusion, and proposes a per-attention-score correction that removes the bias with negligible overhead, recovering near-BF16 video quality at INT2 quantization.
High-accuracy Low-Bit KV-Cache Quantization via Local Distribution Restoration
This paper identifies that low-bit KV cache quantization degrades LLM accuracy due to structured local misranking of logits, and proposes DGAP, a method that restores the local distribution of top-K candidates, recovering RULER accuracy from 47.8% to 83.2% on Llama-3.1-8B with minimal overhead.
Data-driven Video Codec with Implicit Neural Representations
This paper presents a video codec that stores video and audio as weights of a sinusoidal representation network, using knowledge distillation and quantization for compression. Experiments show a 2.61x compression ratio compared to the original network, but quality lags behind standard codecs like H.264 and HEVC.