arXiv subjects

Jianze Li

Publications and source records attributed to Jianze Li.

At least 19 recordsLinked to original sources

arXiv2026

Upper Bounds on the Strassen CP Rank Gap and Symmetric Rank Gap of Third-Order Tensors

For two third-order tensors over $\mathbb{R}$ or $\mathbb{C}$, we define the Strassen CP rank gap as the sum of their CP ranks minus the CP rank of their direct sum. For arbitrary third-order tensors, we derive explicit upper bounds on this gap in terms of CP rank excess and dimension. For symmetric third-order tensors, we similarly define the Strassen symmetric rank gap using symmetric rank and establish explicit upper bounds in terms of symmetric rank excess, CP rank excess, and dimension. These bounds also yield sufficient conditions for the respective gaps to vanish, ensuring additivity of CP rank or symmetric rank under direct sums.

math.RA↗

arXiv2026

DRScaffold: Boosting Dense-Scene Reasoning in Lightweight Vision Language Models

Lightweight vision-language models perform competitively on standard benchmarks yet fail systematically in dense-scene reasoning, where multiple objects, attributes, and relations must be jointly grounded and resolved through multi-step inference. Such capability is critical for real-world applications where models must reliably interpret cluttered environments. Yet existing training signals provide no explicit grounding between reasoning steps and the underlying visual entities and relations, leaving lightweight models free to generate fluent but visually unanchored reasoning chains. To address this gap, we first introduce DRBench, a benchmark of 14,573 questions across 2,943 images, organized into five task categories spanning three progressive reasoning layers. Building on DRBench, we propose DRScaffold, a supervised fine-tuning framework that decomposes the supervision target into four causally ordered stages, promoting grounded reasoning without architectural modification. Experiments on three lightweight VLMs demonstrate substantial gains on DRBench while preserving or improving performance on general-purpose benchmarks. Notably, Qwen2.5-VL-3B trained with DRScaffold surpasses the frozen Qwen2.5-VL-32B under zero-shot direct-answer prompting on DRBench, demonstrating that structured supervision can substitute for a significant portion of model scale in dense-scene reasoning. Our code is available at https://github.com/irene-shi/DRScaffold.

cs.CV↗

arXiv2026

Rank and Border Rank Additivity of Complex Tensors with a Hankel Summand

We prove that the CP rank of a complex Hankel tensor of order $m\ge3$ is additive with that of every complex tensor of the same order. Together with the known equality of CP and symmetric ranks for complex Hankel tensors, this yields symmetric rank additivity whenever the other summand has equal CP and symmetric ranks. We also prove that ordinary and symmetric border ranks coincide and are additive on direct sums of complex Hankel tensors of any common order at least three. Their common value is the sum of the middle binary catalecticant ranks. For order three, we further prove border-rank additivity when the other summand has border rank at most three, or is symmetric with at most five essential variables.

math.RA↗

arXiv2026

SCOPE-VSR: A Broad-Coverage Training Dataset for Video Super-Resolution

Open-domain video super-resolution (VSR) requires temporally coherent details across widely varying scenes and subjects. However, existing training datasets are often semantically imbalanced and insufficiently diverse. We address this data bottleneck with a structured curation pipeline for open-domain VSR, SCOPE. In this pipeline, we combine task-oriented video selection with hierarchical semantic rebalancing. Applying this pipeline to a large-scale collection of data sources, we construct the SCOPE-VSR dataset for training and SCOPE-test for testing. SCOPE-VSR has approximately 27k videos and provides balanced and diverse coverage of various real-world scenarios and subjects. We further develop GraCo-VSR, a novel framework for one-step video restoration. GraCo-VSR adopts a reliability-aware design with explicit supervision to reduce degradation-induced distortions in motion and texture guidance, allowing the model to benefit more effectively from the diverse training content in SCOPE-VSR. Evaluations show that our model achieves strong reconstruction fidelity and perceptual quality. Our code is available at: https://github.com/gyliao2768/SCOPE-VSR.

cs.CV↗

arXiv2026

FrameLite: Decision-Oriented Aesthetic Cropping with a Compact Vision-Language Model

Aesthetic image cropping improves composition by balancing subjects, context, and image boundaries. Large vision-language models bring semantic knowledge to this task, but their scale limits lightweight photography assistance. We propose FrameLite, a compact cropping model built on Qwen3.5-0.8B. Our approach is motivated by the observation that a small model can generate better crops than its first answer suggests. We develop three-stage training to turn this potential into stronger direct predictions. Spatial and composition training establishes localization and framing preferences. Quality-Guided Online Supervision (QOS) then learns the highest-quality complete answers among the current policy's generated candidates. Greedy-Referenced Optimization (GRO) further refines the policy using positive and negative quality differences relative to its current greedy answer. Both online stages use annotation-derived feedback, while inference returns one coordinate answer without search or an external scorer. Experiments show that FrameLite outperforms the compared large-VLM croppers across public benchmarks with a substantially smaller backbone. Controlled studies confirm that selecting high-quality training answers improves the final model beyond online sampling alone. We will release our code and model weights at https://github.com/JianzeLi-114/FrameLite .

cs.CV↗

arXiv2026

Globally Convergent Jacobi-Type Algorithms for Robust Approximate Diagonalization of Higher-Order Cumulant Tensors

The sums of the absolute values of higher-order cumulants are classical contrast functions for independent component analysis (ICA) and lead naturally to robust approximate diagonalization of higher-order cumulant tensors using $\ell_1$-based diagonalization over the orthogonal or unitary group. We propose Jacobi-SubG, a Jacobi-type framework that selects index pairs using Riemannian subdifferential information and performs plane rotations with a refinement step that preserves nonzero diagonal contractions. For the general tensor model, we prove global convergence of the entire iterate sequence from any initialization with nonzero diagonal contractions, provided that the subproblem solvers satisfy a uniform sufficient-ascent estimate. For real and complex third- and fourth-order cumulant tensors, which are widely used in ICA, we develop specialized subproblem solvers satisfying these conditions. The resulting algorithms thus enjoy global convergence under the simple initialization condition that all diagonal contractions are nonzero. Numerical experiments on ICA and tensor diagonalization show that the proposed Jacobi-SubG variants achieve better source recovery and higher final objective values in several settings, particularly at the higher tested noise levels.

math.NA↗

arXiv2026

Globally Convergent Block Coordinate Descent Algorithms for Semi-Nonnegative Independent Component Analysis

We propose a subgradient-based block coordinate descent (BCD-SubG) framework for nonconvex optimization over products of probability simplices and a closed feasible set, motivated by semi-nonnegative independent component analysis (Semi-NNICA). Elementary column transformations are used to preserve simplex feasibility, while first-order information guides block and pair selection. We establish the global convergence of BCD-SubG using a constrained Lojasiewicz inequality; additional eventual-interiority conditions yield stationary limits and convergence rates. We then specialize BCD-SubG to the direct-fit and indirect-fit Semi-NNICA models and derive explicit updates that ensure global convergence without further assumptions. For both models, eventual interiority also yields stationary limits and convergence-rate estimates. Experiments on synthetic mixtures and simulated magnetic resonance spectroscopy data show generally lower recovery errors and shorter wall-clock times than the tested Semi-NNICA baselines. Joint diagonalization experiments further demonstrate lower final objective values.

math.OC↗

arXiv2026

The Comon Rank Gap of a Third-Order Symmetric Tensor Can Exceed One

For a third-order symmetric tensor, the Comon rank gap is the difference between its CP rank and its symmetric rank. We study direct sums of Lovitz's rational $27$-dimensional tensor, whose two ranks are $55$ and $56$. Two copies have ranks $110$ and $112$, and three copies have ranks $165$ and $168$, over both the real and the complex fields. Thus direct sums yield concise cubic tensors on spaces of dimensions $54$ and $81$ with Comon rank gaps two and three.

math.RA↗

arXiv2026

Discrete Annotation, Continuous Preference: Rethinking Supervision for Accurate and Generalizable Aesthetic Image Cropping

Aesthetic image cropping aims to identify the optimal crop of an image in terms of aesthetics and composition. While supervision based on annotated data is fundamental, the field has been hindered by a long-standing problem: existing datasets suffer from (1) human subjectivity and (2) rigid discreteness confined to fixed sampling grids. These flawed annotations not only limit the accuracy and generalization of trained models but also severely distort fair evaluation. To overcome this, we propose to model human cropping preference as a multi-peaked, continuous, and sharp field over the crop space. We introduce the Continuous Preference Field (CPF), which recovers a dense preference landscape from discrete annotations through (1) peak clustering, (2) off-lattice refinement, (3) negative shaping, and (4) field assembly. Based on this, we train CPIC, a VLM-based cropping model optimized via GRPO with the CPF reward, which overcomes template collapse, achieving state-of-the-art performance and exceptional out-of-domain generalization. Finally, to resolve the long-standing benchmark evaluation crisis, we introduce CPICD, a comprehensive recalibration of existing ground-truth boxes. By leveraging the CPF to correct grid-bound artifacts across mainstream benchmarks, CPICD establishes a rigorous and reliable foundation for future cropping research. Extensive experiments and user studies demonstrate the superiority of our CPF, CPIC, and CPICD. Code, model, and data are available at https://github.com/zzqingz/CPIC.

cs.CV↗

arXiv2026

Upper Bounds on the Comon Rank Gap of Real Symmetric Third-Order Tensors

Let $T$ be a nonzero real symmetric third-order tensor. Denote by $r(T)$ its CP rank, by $s(T)$ its symmetric rank, by $d(T)$ its essential dimension, and put $e(T)=r(T)-d(T)$. We establish two complementary upper bounds for the Comon rank gap $$Δ(T):= s(T)-r(T)\leq\min\{C_{ec}(e(T)),\ C_{dc}(d(T))\}.$$ The excess-controlled function $C_{ec}(e)$ is zero for $0\leq e\leq3$ and equals $9e-3$ for $e\geq4$. In particular, the equality $C_{ec}(e(T))=0$ for $0\leq e\leq3$ extends Friedland's flattening-rank criterion for real symmetric third-order tensors by two additional rank layers. The dimension-controlled function $C_{dc}(d)$ is zero for $d=0$, equals $(d^2-4d+10)/3$ for $1\leq d\leq18$, and equals $d^2/4+d/4+3/2$ for $d\geq19$. The excess estimate follows from the alternating equations and decomposable-bivector support geometry; a low-wedge descent gives the vanishing range $e\leq3$. The dimension estimate combines a transverse polar-completion bound with a shifted estimate for the maximum real Waring rank. Thus the two bounds measure different features of the same tensor and remain effective in complementary regimes.

math.RA↗

arXiv2026

SARSteer: Safeguarding Large Audio-Language Models via Safe-Ablated Refusal Steering

Large Audio-Language Models (LALMs) are becoming essential as a powerful multimodal backbone for real-world applications. However, recent studies show that audio inputs can more easily elicit harmful responses than text, exposing new risks toward deployment. While safety alignment has made initial advances in LLMs and Large Vision-Language Models (LVLMs), we find that vanilla adaptation of these approaches to LALMs faces two key limitations: 1) LLM-based steering fails under audio input due to the large distributional gap between activations, and 2) prompt-based defenses induce over-refusals on benign-speech queries. To address these challenges, we propose Safe-Ablated Refusal Steering (SARSteer), the first inference-time defense framework for LALMs. Specifically, SARSteer leverages text-derived refusal steering to enforce rejection without manipulating audio inputs and introduces decomposed safe-space ablation to mitigate over-refusal. Extensive experiments demonstrate that SARSteer significantly improves harmful-query refusal while preserving benign responses, establishing a principled step toward safety alignment in LALMs. The codes and constructed datasets are released at https://github.com/linweiii/SARSteer.

cs.SD↗

arXiv2026

RedEdit: Agentic Red-Teaming of Image Safety Classifiers via MCTS-Guided Photo-Editing

Image safety classifiers serve as a critical component of contemporary content moderation systems on the internet. However, their resilience against user-style malicious image editing remains underexplored. Such behaviors are highly prevalent in daily scenarios but difficult to fully reproduce. To explore this vulnerability, we introduce RedEdit, a novel black-box red-teaming agent that formulates photo-editing evasion as a combinatorial search problem over edit-tool sequences. It adopts a Vision-Language-Model (VLM)-based proposer to generate semantically targeted candidate edits and a Monte Carlo Tree Search (MCTS) planner to prioritize promising edit paths while backtracking from ineffective ones. Together, the proposer and planner instantiate two key capabilities of human attackers, i.e., domain knowledge and iterative backtracking, respectively, to reproduce this practical threat. Our extensive experiments on UnsafeBench reveal profound systemic vulnerabilities: fewer than two edits on average enable 76.2% of unsafe images to evade detectors, while retaining 93.0% malicious semantics, meaning that such manipulated content remains perceptually malicious to humans while easily bypassing automated moderation. We therefore appeal to the community for more attention to this overlooked practical threat.

cs.CR↗

arXiv2026

The First Challenge on Remote Sensing Infrared Image Super-Resolution at NTIRE 2026: Benchmark Results and Method Overview

This paper presents the NTIRE 2026 Remote Sensing Infrared Image Super-Resolution (x4) Challenge, one of the associated challenges of NTIRE 2026. The challenge aims to recover high-resolution (HR) infrared images from low-resolution (LR) inputs generated through bicubic downsampling with a x4 scaling factor. The objective is to develop effective models or solutions that achieve state-of-the-art performance for infrared image SR in remote sensing scenarios. To reflect the characteristics of infrared data and practical application needs, the challenge adopts a single-track setting. A total of 115 participants registered for the competition, with 13 teams submitting valid entries. This report summarizes the challenge design, dataset, evaluation protocol, main results, and the representative methods of each team. The challenge serves as a benchmark to advance research in infrared image super-resolution and promote the development of effective solutions for real-world remote sensing applications.

cs.CV↗

arXiv2026

The Fourth Challenge on Image Super-Resolution ($\times$4) at NTIRE 2026: Benchmark Results and Method Overview

This paper presents the NTIRE 2026 image super-resolution ($\times$4) challenge, one of the associated competitions of the NTIRE 2026 Workshop at CVPR 2026. The challenge aims to reconstruct high-resolution (HR) images from low-resolution (LR) inputs generated through bicubic downsampling with a $\times$4 scaling factor. The objective is to develop effective super-resolution solutions and analyze recent advances in the field. To reflect the evolving objectives of image super-resolution, the challenge includes two tracks: (1) a restoration track, which emphasizes pixel-wise fidelity and ranks submissions based on PSNR; and (2) a perceptual track, which focuses on visual realism and evaluates results using a perceptual score. A total of 194 participants registered for the challenge, with 31 teams submitting valid entries. This report summarizes the challenge design, datasets, evaluation protocol, main results, and methods of participating teams. The challenge provides a unified benchmark and offers insights into current progress and future directions in image super-resolution.

cs.CV↗

arXiv2025

Convergence analysis of the transformed gradient projection algorithms on compact matrix manifolds

In this paper, we study the optimization problem on a compact matrix manifold. While existing feasible algorithms can be broadly categorized into retraction-based and projection-based methods, compared to the more comprehensive and in-depth algorithmic and convergence research framework for retraction-based line-search (RetrLS) algorithms using only tangent vectors, the theoretical understanding and algorithmic design of projection-based line-search (ProjLS) algorithms remain limited, especially when general search directions and stepsizes are involved. To bridge this gap, we propose a unified algorithmic framework called the Transformed Gradient Projection (TGP) algorithm. The key idea is to construct the search direction as a transformed Riemannian (or Euclidean) gradient augmented by an additional normal component, allowing the framework to encompass and generalize numerous existing algorithms. Then, we conduct a thorough exploration of the convergence properties of the TGP algorithms under various stepsizes, including the Armijo, Zhang-Hager type nonmonotone Armijo, and fixed stepsizes. To achieve this, we extensively analyze the geometric properties of the projection onto compact matrix manifolds, which may be of independent interest. Building upon these insights, we establish the weak convergence, iteration complexity, and global convergence of TGP algorithms under three distinct stepsizes. In cases where the compact matrix manifold is the Stiefel or Grassmann manifold, our convergence results either encompass or surpass those found in the literature. Finally, through a series of numerical experiments and theoretical analysis, we observe that different choices of scaling matrices and normal components in the search direction of TGP algorithms can lead to significantly different performance in practice.

math.OC↗

arXiv2025

Asymmetric VAE for One-Step Video Super-Resolution Acceleration

Diffusion models have significant advantages in the field of real-world video super-resolution and have demonstrated strong performance in past research. In recent diffusion-based video super-resolution (VSR) models, the number of sampling steps has been reduced to just one, yet there remains significant room for further optimization in inference efficiency. In this paper, we propose FastVSR, which achieves substantial reductions in computational cost by implementing a high compression VAE (spatial compression ratio of 16, denoted as f16). We design the structure of the f16 VAE and introduce a stable training framework. We employ pixel shuffle and channel replication to achieve additional upsampling. Furthermore, we propose a lower-bound-guided training strategy, which introduces a simpler training objective as a lower bound for the VAE's performance. It makes the training process more stable and easier to converge. Experimental results show that FastVSR achieves speedups of 111.9 times compared to multi-step models and 3.92 times compared to existing one-step models. We will release code and models at https://github.com/JianzeLi-114/FastVSR.

cs.CV↗

arXiv2025

BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model

Backdoor learning is a critical research topic for understanding the vulnerabilities of deep neural networks. While the diffusion model (DM) has been broadly deployed in public over the past few years, the understanding of its backdoor vulnerability is still in its infancy compared to the extensive studies in discriminative models. Recently, many different backdoor attack and defense methods have been proposed for DMs, but a comprehensive benchmark for backdoor learning on DMs is still lacking. This absence makes it difficult to conduct fair comparisons and thorough evaluations of the existing approaches, thus hindering future research progress. To address this issue, we propose \textit{BackdoorDM}, the first comprehensive benchmark designed for backdoor learning on DMs. It comprises nine state-of-the-art (SOTA) attack methods, four SOTA defense strategies, and three useful visualization analysis tools. We first systematically classify and formulate the existing literature in a unified framework, focusing on three different backdoor attack types and five backdoor target types, which are restricted to a single type in discriminative models. Then, we systematically summarize the evaluation metrics for each type and propose a unified backdoor evaluation method based on multimodal large language model (MLLM). Finally, we conduct a comprehensive evaluation and highlight several important conclusions. We believe that BackdoorDM will help overcome current barriers and contribute to building a trustworthy artificial intelligence generated content (AIGC) community. The codes are released in https://github.com/linweiii/BackdoorDM.

cs.CR↗

arXiv2025

Unleashing the Power of One-Step Diffusion based Image Super-Resolution via a Large-Scale Diffusion Discriminator

Diffusion models have demonstrated excellent performance for real-world image super-resolution (Real-ISR), albeit at high computational costs. Most existing methods are trying to derive one-step diffusion models from multi-step counterparts through knowledge distillation (KD) or variational score distillation (VSD). However, these methods are limited by the capabilities of the teacher model, especially if the teacher model itself is not sufficiently strong. To tackle these issues, we propose a new One-Step \textbf{D}iffusion model with a larger-scale \textbf{D}iffusion \textbf{D}iscriminator for SR, called D$^3$SR. Our discriminator is able to distill noisy features from any time step of diffusion models in the latent space. In this way, our diffusion discriminator breaks through the potential limitations imposed by the presence of a teacher model. Additionally, we improve the perceptual loss with edge-aware DISTS (EA-DISTS) to enhance the model's ability to generate fine details. Our experiments demonstrate that, compared with previous diffusion-based methods requiring dozens or even hundreds of steps, our D$^3$SR attains comparable or even superior results in both quantitative metrics and qualitative evaluations. Moreover, compared with other methods, D$^3$SR achieves at least $3\times$ faster inference speed and reduces parameters by at least 30\%. We will release code and models at https://github.com/JianzeLi-114/D3SR.

cs.CV↗