Newest papers

Training-free Acceleration for Blind Compressed Image Restoration via Activation-Calibrated Singular Value Decomposition

姚苏杭, Xin Li, Xijun Wang, Yufei Wang, Yeying Jin, Zhibo Chen · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29959 · cited by 0 works (OpenAlex) · open-access PDF

Russian

Трансформерные активные модели, основанные на ошибках 216, характеризуют сильную производительность при восстановлении слепого сжатого изображения (CIR) при неизвестных настройках сжатия, но их плотно сложенные операторы проекции делают развертывание дорогостоящим. Мы изучаем беспрерывное ускорение для таких моделей с помощью эмпирического анализа на уровне оператора. В частности, мы профилируем, где вычисление концентрируется, характеризуем распределения с низким уровнем активации по каналу под сжатыми входами, и изучаем деградацию SVD (TSVD) в виде однозначного распада ценностей представительных весов проекции. Эти измерения идентифицируют повторяющиеся проекции в внимании, питании вперед, конволюцион

Machine translation (Meta M2M100); the English original follows.

Abstract

Transformer-based models achieve strong performance on blind compressed image restoration (CIR) under unknown compression settings, but their densely stacked projection operators make deployment costly. We study training-free acceleration for such models through an empirical, operator-level analysis. Specifically, we profile where computation is concentrated, characterize channel-wise activation distributions under compressed inputs, and examine the singular-value decay of representative projection weights. These measurements identify the recurring projections in the attention, feed-forward, convolutional, and prompt-conditioning pathways as effective low-rank replacement targets that preserve the original inference graph. They further establish truncated SVD (TSVD) as a principled training-free baseline for blind CIR and reveal why a uniform low-rank objective is mismatched to the channel-imbalanced feature distributions of compressed images. Accordingly, we formulate Activation-Calibrated Singular Value Decomposition (ASVD) for this restoration setting, deriving low-rank factors entirely offline from pretrained weights and lightweight activation statistics gathered from only a few compressed calibration images. By weighting the decomposition with channel-wise activation magnitudes, our ASVD assigns larger penalties to errors on more active channels, requiring neither retraining nor online calibration and leaving the restoration pipeline intact. When instantiated on PromptCIR at rank 64, ASVD consistently improves PSNR-B over TSVD across four standard benchmarks under identical parameter and MAC budgets. At this operating point, the factorized model cuts parameters by 59% (34.75M to 14.34M) and computation by 16.1% (258.79 to 217.16 GMACs for a 256 x 256 input), enabling acceleration without retraining.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Singular value decomposition, Projection (relational algebra), Computation, Acceleration, Pipeline (software), Compressed sensing

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange