Papers

Automated visual inspection of locomotive coil springs presents significant challenges due to the morphological diversity of surface defects, substantial scale variations, and complex industrial backgrounds. This paper proposes MSD-DETR (Multi-Scale Deformable Detection Transformer), a novel detection framework that addresses these challenges through three key innovations: (1) a structural re-parameterization strategy that decouples training-time multi-branch topology from inference-time efficiency, enhancing feature extraction while maintaining real-time performance; (2) a deformable attention mechanism that enables content-adaptive spatial sampling, allowing dynamic focus on defect-relevant regions regardless of morphological irregularity; and (3) a cross-scale feature fusion architecture incorporating GSConv modules and VoVGSCSP blocks for effective multi-resolution information aggregation. Comprehensive experiments on a real-world locomotive coil spring dataset demonstrate that MSD-DETR achieves 92.4% mAP@0.5 at 98 FPS, outperforming state-of-the-art detectors including YOLOv8 (+3.1% mAP) and the baseline RT-DETR (+2.8% mAP) while maintaining comparable inference speed, establishing a new benchmark for industrial coil spring quality inspection.

In different multimodal scenarios, it needs to integrate and utilize information across modalities in a specific way based on the demands of the task. Different integration ways between modalities are referred to as "multimodal interaction". How well a model handles various multimodal interactions largely characterizes its multimodal ability. In this paper, we introduce MIBench, a comprehensive benchmark designed to evaluate the multimodal interaction capabilities of Large Multimodal Models (LMMs), which formulates each instance as a (con_v , con_t, task) triplet with contexts from vision and text, necessitating that LMMs employ correct forms of multimodal interaction to effectively complete the task. MIBench assesses models from three key aspects: the ability to source information from vision-centric or text-centric cues, and the ability to generate new information from their joint synergy. Each interaction capability is evaluated hierarchically across three cognitive levels: Recognition, Understanding, and Reasoning. MIBench comprises over 10,000 vision-text context pairs spanning 32 distinct tasks. Evaluation of state-of-the-art LMMs show that: (1) LMMs' ability on multimodal interaction remains constrained, despite the scaling of model parameters and training data; (2) they are easily distracted by textual modalities when processing vision information; (3) they mostly possess a basic capacity for multimodal synergy; and (4) natively trained multimodal models show noticeable deficits in fundamental interaction ability. We expect that these observations can serve as a reference for developing LMMs with more enhanced multimodal ability in the future.

Full-waveform inversion (FWI) is a high-resolution seismic imaging method that estimates subsurface velocity by matching simulated and recorded waveforms. However, FWI is highly nonlinear, prone to cycle skipping, and sensitive to noise, particularly when low frequencies are missing or the initial model is poor, leading to failures under imperfect acquisition. Diffusion-regularized FWI introduces generative priors to encourage geologically realistic models, but these priors typically require costly offline pretraining and can deteriorate under distribution shift. Moreover, they assume Gaussian initialization and a fixed noise schedule, in which it is unclear how to map a deterministic FWI iterate and its starting model to a well-defined diffusion time or noise level. To address these limitations, we introduce Self-Flow-Matching assisted Full-Waveform Inversion (SFM-FWI), a physics-driven framework that eliminates the need for large-scale offline pretraining while avoiding the noise-level alignment ambiguity. SFM-FWI leverages flow matching to learn a transport field without assuming Gaussian initialization or a predefined noise schedule, so the initial model can be used directly as the starting point of the dynamics. Our approach trains a single flow network online using the governing physics and observed data. At each outer iteration, we build an interpolated model and update the flow by backpropagating the FWI data misfit, providing self-supervision without external training pairs. Experiments on challenging synthetic benchmarks show that SFM-FWI delivers more accurate reconstructions, greater noise robustness, and more stable convergence than standard FWI and pretraining-free regularization methods.

Recent progress in text-conditioned human motion generation has been largely driven by diffusion models trained on large-scale human motion data. Building on this progress, recent methods attempt to transfer such models for character animation and real robot control by applying a Whole-Body Controller (WBC) that converts diffusion-generated motions into executable trajectories. While WBC trajectories become compliant with physics, they may expose substantial deviations from original motion. To address this issue, we here propose PhysMoDPO, a Direct Preference Optimization framework. Unlike prior work that relies on hand-crafted physics-aware heuristics such as foot-sliding penalties, we integrate WBC into our training pipeline and optimize diffusion model such that the output of WBC becomes compliant both with physics and original text instructions. To train PhysMoDPO we deploy physics-based and task-specific rewards and use them to assign preference to synthesized trajectories. Our extensive experiments on text-to-motion and spatial control tasks demonstrate consistent improvements of PhysMoDPO in both physical realism and task-related metrics on simulated robots. Moreover, we demonstrate that PhysMoDPO results in significant improvements when applied to zero-shot motion transfer in simulation and for real-world deployment on a G1 humanoid robot.

Machine learning approaches to spatiotemporal physical systems have primarily focused on next-frame prediction, with the goal of learning an accurate emulator for the system's evolution in time. However, these emulators are computationally expensive to train and are subject to performance pitfalls, such as compounding errors during autoregressive rollout. In this work, we take a different perspective and look at scientific tasks further downstream of predicting the next frame, such as estimation of a system's governing physical parameters. Accuracy on these tasks offers a uniquely quantifiable glimpse into the physical relevance of the representations of these models. We evaluate the effectiveness of general-purpose self-supervised methods in learning physics-grounded representations that are useful for downstream scientific tasks. Surprisingly, we find that not all methods designed for physical modeling outperform generic self-supervised learning methods on these tasks, and methods that learn in the latent space (e.g., joint embedding predictive architectures, or JEPAs) outperform those optimizing pixel-level prediction objectives. Code is available at https://github.com/helenqu/physical-representation-learning.

Vision-to-code tasks require models to reconstruct structured visual inputs, such as charts, tables, and SVGs, into executable or structured representations with high visual fidelity. While recent Large Vision Language Models (LVLMs) achieve strong results via supervised fine-tuning, reinforcement learning remains challenging due to misaligned reward signals. Existing rewards either rely on textual rules or coarse visual embedding similarity, both of which fail to capture fine-grained visual discrepancies and are vulnerable to reward hacking. We propose Visual Equivalence Reward Model (Visual-ERM), a multimodal generative reward model that provides fine-grained, interpretable, and task-agnostic feedback to evaluate vision-to-code quality directly in the rendered visual space. Integrated into RL, Visual-ERM improves Qwen3-VL-8B-Instruct by +8.4 on chart-to-code and yields consistent gains on table and SVG parsing (+2.7, +4.1 on average), and further strengthens test-time scaling via reflection and revision. We also introduce VisualCritic-RewardBench (VC-RewardBench), a benchmark for judging fine-grained image-to-image discrepancies on structured visual data, where Visual-ERM at 8B decisively outperforms Qwen3-VL-235B-Instruct and approaches leading closed-source models. Our results suggest that fine-grained visual reward supervision is both necessary and sufficient for vision-to-code RL, regardless of task specificity.

Evolutions in the world, such as water pouring or ice melting, happen regardless of being observed. Video world models generate "worlds" via 2D frame observations. Can these generated "worlds" evolve regardless of observation? To probe this question, we design a benchmark to evaluate whether video world models can decouple state evolution from observation. Our benchmark, STEVO-Bench, applies observation control to evolving processes via instructions of occluder insertion, turning off the light, or specifying camera "lookaway" trajectories. By evaluating video models with and without camera control for a diverse set of naturally-occurring evolutions, we expose their limitations in decoupling state evolution from observation. STEVO-Bench proposes an evaluation protocol to automatically detect and disentangle failure modes of video world models across key aspects of natural state evolution. Analysis of STEVO-Bench results provide new insight into potential data and architecture bias of present-day video world models. Project website: https://glab-caltech.github.io/STEVOBench/. Blog: https://ziqi-ma.github.io/blog/2026/outofsight/

Spatio-temporal scene graphs provide a principled representation for modeling evolving object interactions, yet existing methods remain fundamentally frame-centric: they reason only about currently visible objects, discard entities upon occlusion, and operate in 2D. To address this, we first introduce ActionGenome4D, a dataset that upgrades Action Genome videos into 4D scenes via feed-forward 3D reconstruction, world-frame oriented bounding boxes for every object involved in actions, and dense relationship annotations including for objects that are temporarily unobserved due to occlusion or camera motion. Building on this data, we formalize World Scene Graph Generation (WSGG), the task of constructing a world scene graph at each timestamp that encompasses all interacting objects in the scene, both observed and unobserved. We then propose three complementary methods, each exploring a different inductive bias for reasoning about unobserved objects: PWG (Persistent World Graph), which implements object permanence via a zero-order feature buffer; MWAE (Masked World Auto-Encoder), which reframes unobserved-object reasoning as masked completion with cross-view associative retrieval; and 4DST (4D Scene Transformer), which replaces the static buffer with differentiable per-object temporal attention enriched by 3D motion and camera-pose features. We further design and evaluate the performance of strong open-source Vision-Language Models on the WSGG task via a suite of Graph RAG-based approaches, establishing baselines for unlocalized relationship prediction. WSGG thus advances video scene understanding toward world-centric, temporally persistent, and interpretable scene reasoning.

Brain tumor classification from magnetic resonance imaging, which is also known as MRI, plays a sensitive role in computer-assisted diagnosis systems. In recent years, deep learning models have achieved high classification accuracy. However, their sensitivity to adversarial perturbations has become an important reliability concern in medical applications. This study suggests a robust brain tumor classification framework that combines Non-Negative Matrix Factorization (NNMF or NMF), lightweight convolutional neural networks (CNNs), and diffusion-based feature purification. Initially, MRI images are preprocessed and converted into a non-negative data matrix, from which compact and interpretable NNMF feature representations are extracted. Statistical metrics, including AUC, Cohen's d, and p-values, are used to rank and choose the most discriminative components. Then, a lightweight CNN classifier is trained directly on the selected feature groups. To improve adversarial robustness, a diffusion-based feature-space purification module is introduced. A forward noise method followed by a learned denoiser network is used before classification. System performance is estimated using both clean accuracy and robust accuracy under powerful adversarial attacks created by AutoAttack. The experimental results show that the proposed framework achieves competitive classification performance while significantly enhancing robustness against adversarial perturbations.The findings presuppose that combining interpretable NNMF-based representations with a lightweight deep approach and diffusion-based defense technique supplies an effective and reliable solution for medical image classification under adversarial conditions.

In modern human-robot collaboration (HRC) applications, multiple perception modules jointly extract visual, auditory, and contextual cues to achieve comprehensive scene understanding, enabling the robot to provide appropriate assistance to human agents intelligently. While executing multiple perception modules on a frame-by-frame basis enhances perception quality in offline settings, it inevitably accumulates latency, leading to a substantial decline in system performance in streaming perception scenarios. Recent work in scene understanding, termed Relevance, has established a solid foundation for developing efficient methodologies in HRC. However, modern perception pipelines still face challenges related to information redundancy and suboptimal allocation of computational resources. Drawing inspiration from the Relevance concept and the information sparsity in HRC events, we propose a novel lightweight perception scheduling framework that efficiently leverages output from previous frames to estimate and schedule necessary perception modules in real-time based on scene context. The experimental results demonstrate that the proposed perception scheduling framework effectively reduces computational latency by up to 27.52% compared to conventional parallel perception pipelines, while also achieving a 72.73% improvement in MMPose activation recall. Additionally, the framework demonstrates high keyframe accuracy, achieving rates of up to 98%. The results validate the framework's capability to enhance real-time perception efficiency without significantly compromising accuracy. The framework shows potential as a scalable and systematic solution for multimodal streaming perception systems in HRC.

Concept Bottleneck Models (CBMs) are interpretable models that route predictions through a layer of human-interpretable concepts. While widely studied in vision and, more recently, in NLP, CBMs remain largely unexplored in multimodal settings. For their explanations to be faithful, CBMs must satisfy two conditions: concepts must be properly detected, and concept representations must encode only their intended semantics, without smuggling extraneous task-relevant or inter-concept information into final predictions, a phenomenon known as leakage. Existing approaches treat concept detection and leakage mitigation as separate problems, and typically improve one at the expense of predictive accuracy. In this work, we introduce f-CBM, a faithful multimodal CBM framework built on a vision-language backbone that jointly targets both aspects through two complementary strategies: a differentiable leakage loss to mitigate leakage, and a Kolmogorov-Arnold Network prediction head that provides sufficient expressiveness to improve concept detection. Experiments demonstrate that f-CBM achieves the best trade-off between task accuracy, concept detection, and leakage reduction, while applying seamlessly to both image and text or text-only datasets, making it versatile across modalities.

Diffusion-based image compression has recently shown outstanding perceptual fidelity, yet its practicality is hindered by prohibitive sampling overhead and high memory usage. Most existing diffusion codecs employ U-Net architectures, where hierarchical downsampling forces diffusion to operate in shallow latent spaces (typically with only 8x spatial downscaling), resulting in excessive computation. In contrast, conventional VAE-based codecs work in much deeper latent domains (16x - 64x downscaled), motivating a key question: Can diffusion operate effectively in such compact latent spaces without compromising reconstruction quality? To address this, we introduce DiT-IC, an Aligned Diffusion Transformer for Image Compression, which replaces the U-Net with a Diffusion Transformer capable of performing diffusion in latent space entirely at 32x downscaled resolution. DiT-IC adapts a pretrained text-to-image multi-step DiT into a single-step reconstruction model through three key alignment mechanisms: (1) a variance-guided reconstruction flow that adapts denoising strength to latent uncertainty for efficient reconstruction; (2) a self-distillation alignment that enforces consistency with encoder-defined latent geometry to enable one-step diffusion; and (3) a latent-conditioned guidance that replaces text prompts with semantically aligned latent conditions, enabling text-free inference. With these designs, DiT-IC achieves state-of-the-art perceptual quality while offering up to 30x faster decoding and drastically lower memory usage than existing diffusion-based codecs. Remarkably, it can reconstruct 2048x2048 images on a 16 GB laptop GPU.

Face de-identification (FDeID) aims to remove personally identifiable information from facial images while preserving task-relevant utility attributes such as age, gender, and expression. It is critical for privacy-preserving computer vision, yet the field suffers from fragmented implementations, inconsistent evaluation protocols, and incomparable results across studies. These challenges stem from the inherent complexity of the task: FDeID spans multiple downstream applications (e.g., age estimation, gender recognition, expression analysis) and requires evaluation across three dimensions (e.g., privacy protection, utility preservation, and visual quality), making existing codebases difficult to use and extend. To address these issues, we present FDeID-Toolbox, a comprehensive toolbox designed for reproducible FDeID research. Our toolbox features a modular architecture comprising four core components: (1) standardized data loaders for mainstream benchmark datasets, (2) unified method implementations spanning classical approaches to SOTA generative models, (3) flexible inference pipelines, and (4) systematic evaluation protocols covering privacy, utility, and quality metrics. Through experiments, we demonstrate that FDeID-Toolbox enables fair and reproducible comparison of diverse FDeID methods under consistent conditions.

Camera motion is a fundamental geometric signal that shapes visual perception and cinematic style, yet current video-capable vision-language models (VideoLLMs) rarely represent it explicitly and often fail on fine-grained motion primitives. We address this gap with a framework of $\textbf{benchmarking}$, $\textbf{diagnosis}$, and $\textbf{injection}$. We curate $\textbf{CameraMotionDataset}$, a large-scale synthetic dataset with explicit camera control, formulate camera motion as constraint-aware multi-label recognition, and construct a VQA benchmark–$\textbf{CameraMotionVQA}$. Across diverse off-the-shelf VideoLLMs, we observe substantial errors in recognizing camera motion primitives. Probing experiments on a Qwen2.5-VL vision encoder suggest that camera motion cues are weakly represented, especially in deeper ViT blocks, helping explain the observed failure modes. To bridge this gap without costly training or fine-tuning, we propose a lightweight, model-agnostic pipeline that extracts geometric camera cues from 3D foundation models (3DFMs), predicts constrained motion primitives with a temporal classifier, and injects them into downstream VideoLLM inference via structured prompting. Experiments demonstrate improved motion recognition and more camera-aware model responses, highlighting geometry-driven cue extraction and structured prompting as practical steps toward a camera-aware VideoLLM and VLA system. The dataset and benchmark is publicly available at https://hf.co/datasets/fengyee/camera-motion-dataset-and-benchmark.

This paper presents NOIR, a framework that reframes core medical imaging tasks as operator learning between continuous function spaces, challenging the prevailing paradigm of discrete grid-based deep learning. Instead of operating on fixed pixel or voxel grids, NOIR embeds discrete medical signals into shared Implicit Neural Representations and learns a Neural Operator that maps between their latent modulations, enabling resolution-independent function-to-function transformations. We evaluate NOIR across multiple 2D and 3D downstream tasks, including segmentation, shape completion, image-to-image translation, and image synthesis, on several public datasets such as Shenzhen, OASIS-4, SkullBreak, fastMRI, as well as an in-house clinical dataset. It achieves competitive performance at native resolution while demonstrating strong robustness to unseen discretizations, and empirically satisfies key theoretical properties of neural operators. The project page is available here: https://github.com/Sidaty1/NOIR-io.

Panoramic imagery provides holistic 360{\deg} visual coverage for environmental perception in quadruped robots. However, existing occupancy prediction methods are primarily designed for wheeled autonomous driving and rely heavily on RGB cues, which limits their robustness in complex, dynamically changing environments. To bridge this gap, we introduce PanoMMOcc, the first real-world panoramic multimodal occupancy dataset for quadruped robots, comprising four sensing modalities collected across diverse scenes. We further propose VoxelHound, a panoramic multimodal occupancy perception framework tailored to legged locomotion and spherical imaging. VoxelHound incorporates a Vertical Jitter Compensation (VJC) module to mitigate severe viewpoint perturbations caused by body pitch and roll during locomotion, enabling more consistent spatial reasoning, and a Multimodal Information Prompt Fusion (MIPF) module to effectively integrate panoramic visual cues with auxiliary modalities for enhanced volumetric occupancy prediction. We also establish a comprehensive benchmark on PanoMMOcc and provide detailed dataset analyses to enable systematic evaluation in challenging embodied perception scenarios. Extensive experiments demonstrate that VoxelHound achieves state-of-the-art performance on PanoMMOcc, with a +4.16 gain in mIoU. The dataset and code will be publicly released to facilitate future research on panoramic multimodal 3D perception for embodied robotic systems at https://github.com/SXDR/PanoMMOcc.

Predicting the manufacturability of CAD designs early, in terms of both feasibility and required effort, is a key goal of Design for Manufacturing (DFM). Despite advances in deep learning for CAD and its widespread use in manufacturing process selection, learning-based approaches for predicting manufacturability within a specific process remain limited. Two key challenges limit progress: inconsistency across prior work in how manufacturability is defined and consequently in the associated learning targets, and a scarcity of suitable datasets. Existing labels vary significantly: they may reflect intrinsic design constraints or depend on specific manufacturing capabilities (such as available tools), and they range from discrete feasibility checks to continuous complexity measures. Furthermore, industrial datasets typically contain only manufacturable parts, offering little signal for infeasible cases, while existing synthetic datasets focus on simple geometries and subtractive processes. To address these gaps, we propose a taxonomy of manufacturability metrics along the axes of configuration dependence and measurement type, allowing clearer scoping of generalizability and learning objectives. Next, we introduce BenDFM, the first synthetic dataset for manufacturability assessment in sheet metal bending. BenDFM contains 20,000 parts, both manufacturable and unmanufacturable, generated with process-aware bending simulations, providing both folded and unfolded geometries and multiple manufacturability labels across the taxonomy, enabling systematic study of previously unexplored learning-based DFM challenges. We benchmark two state-of-the-art 3D learning architectures on BenDFM, showing that graph-based representations that capture relationships between part surfaces achieve better accuracy, and that predicting metrics that depend on specific manufacturing setups remains more challenging.

We introduce CRYSTAL (Clear Reasoning via Yielded Steps, Traceability, and Logic), a diagnostic benchmark with 6,372 instances that evaluates multimodal reasoning through verifiable intermediate steps. We propose two complementary metrics: Match F1, which scores step-level precision and recall via semantic similarity matching, and Ordered Match F1, which further penalizes disordered reasoning chains. References are constructed through a Delphi-inspired pipeline in which four independent MLLMs generate trajectories, which are then aggregated via semantic clustering and validated through human quality gates. Evaluation of 20 MLLMs, including commercial frontier systems not used during benchmark construction, reveals systematic failures that are invisible to answer accuracy: universal cherry-picking (precision far exceeds recall), non-monotonic scaling trade-offs, and disordered reasoning in which no competitive model preserves more than 60% of matched steps in the correct order. Beyond evaluation, we propose the Causal Process Reward (CPR), a multiplicative reward that couples answer correctness with step-level alignment, and CPR-Curriculum, which progressively increases reasoning difficulty during training. CPR-Curriculum achieves a 32% improvement in Match F1 via GRPO where additive reward strategies fail, improving reasoning without manual step annotation.

We introduce SldprtNet, a large-scale dataset comprising over 242,000 industrial parts, designed for semantic-driven CAD modeling, geometric deep learning, and the training and fine-tuning of multimodal models for 3D design. The dataset provides 3D models in both .step and .sldprt formats to support diverse training and testing. To enable parametric modeling and facilitate dataset scalability, we developed supporting tools, an encoder and a decoder, which support 13 types of CAD commands and enable lossless transformation between 3D models and a structured text representation. Additionally, each sample is paired with a composite image created by merging seven rendered views from different viewpoints of the 3D model, effectively reducing input token length and accelerating inference. By combining this image with the parameterized text output from the encoder, we employ the lightweight multimodal language model Qwen2.5-VL-7B to generate a natural language description of each part's appearance and functionality. To ensure accuracy, we manually verified and aligned the generated descriptions, rendered images, and 3D models. These descriptions, along with the parameterized modeling scripts, rendered images, and 3D model files, are fully aligned to construct SldprtNet. To assess its effectiveness, we fine-tuned baseline models on a dataset subset, comparing image-plus-text inputs with text-only inputs. Results confirm the necessity and value of multimodal datasets for CAD generation. It features carefully selected real-world industrial parts, supporting tools for scalable dataset expansion, diverse modalities, and ensured diversity in model complexity and geometric features, making it a comprehensive multimodal dataset built for semantic-driven CAD modeling and cross-modal learning.

The growing interest in embodied agents increases the demand for spatiotemporal video understanding, yet existing benchmarks largely emphasize extractive reasoning, where answers can be explicitly presented within spatiotemporal events. It remains unclear whether multimodal large language models can instead perform abstractive spatiotemporal reasoning, which requires integrating observations over time, combining dispersed cues, and inferring implicit spatial and contextual structure. To address this gap, we formalize abstractive spatiotemporal reasoning from videos by introducing a structured evaluation taxonomy that systematically targets its core dimensions and constructs a controllable, scenario-driven synthetic egocentric video dataset tailored to evaluate abstractive spatiotemporal reasoning capabilities, spanning object-, room-, and floor-plan-level scenarios. Based on this framework, we present VAEX-BENCH, a benchmark comprising five abstractive reasoning tasks together with their extractive counterparts. Our extensive experiments compare the performance of state-of-the-art MLLMs under extractive and abstractive settings, exposing their limitations on abstractive tasks and providing a fine-grained analysis of the underlying bottlenecks. The dataset will be released soon.

Large-scale video generative models are trained on vast and diverse visual data, enabling them to internalize rich structural, semantic, and dynamic priors of the visual world. While these models have demonstrated impressive generative capability, their potential as general-purpose visual learners remains largely untapped. In this work, we introduce V-Bridge, a framework that bridges this latent capacity to versatile few-shot image restoration tasks. We reinterpret image restoration not as a static regression problem, but as a progressive generative process, and leverage video models to simulate the gradual refinement from degraded inputs to high-fidelity outputs. Surprisingly, with only 1,000 multi-task training samples (less than 2% of existing restoration methods), pretrained video models can be induced to perform competitive image restoration, achieving multiple tasks with a single model, rivaling specialized architectures designed explicitly for this purpose. Our findings reveal that video generative models implicitly learn powerful and transferable restoration priors that can be activated with only extremely limited data, challenging the traditional boundary between generative modeling and low-level vision, and opening a new design paradigm for foundation models in visual tasks.

Understanding whether attention mechanisms converge to the kernel regime is foundational to the validity of influence functions for transformer accountability. Exact NTK characterization of softmax attention is precluded by its exponential nonlinearity; linearized attention is the canonical tractable proxy and the object of study here. This paper establishes that even this proxy does not converge to its NTK limit at any practical width, revealing a fundamental trade-off in the learning dynamics of attention. An exact correspondence is established between parameter-free linearized attention and a data-dependent Gram-induced kernel; spectral amplification analysis shows that the attention transformation cubes the Gram matrix's condition number, requiring width $m = \Omega(\kappa_d(\mathbf{G})^6 n\log n)$ for NTK convergence, where $\kappa_d(\mathbf{G})$ is the effective condition number of the rank-$\min(n,d)$ truncation of the input Gram matrix; for natural image datasets this threshold is physically infeasible ($m \gg 10^{24}$ for MNIST and $m \gg 10^{29}$ for CIFAR-10, 12–17 orders of magnitude beyond the largest known architectures). Influence malleability is introduced to characterize this non-convergence: linearized attention exhibits 2–9$\times$ higher malleability than ReLU networks under adversarial data perturbation, with the gap depending on dataset condition number and task setting. A dual implication is established: the same data-dependent kernel is shown theoretically to reduce approximation error when targets align with the data geometry, while, empirically, creating vulnerability to adversarial manipulation of the training data. The structural argument extends to trainable QKV attention under standard initialization, with direct consequences for influence methods applied to deployed transformer architectures.

Text-guided 3D motion editing has seen success in single-person scenarios, but its extension to multi-person settings is less explored due to limited paired data and the complexity of inter-person interactions. We introduce the task of multi-person 3D motion editing, where a target motion is generated from a source and a text instruction. To support this, we propose InterEdit3D, a new dataset with manual two-person motion change annotations, and a Text-guided Multi-human Motion Editing (TMME) benchmark. We present InterEdit, a synchronized classifier-free conditional diffusion model for TMME. It introduces Semantic-Aware Plan Token Alignment with learnable tokens to capture high-level interaction cues and an Interaction-Aware Frequency Token Alignment strategy using DCT and energy pooling to model periodic motion dynamics. Experiments show that InterEdit improves text-to-motion consistency and edit fidelity, achieving state-of-the-art TMME performance. The dataset and code will be released at https://github.com/YNG916/InterEdit.

Real-time rooftop wind-speed distribution is important for the safe operation of drones and urban air mobility systems, wind control systems, and rooftop utilization. However, rooftop flows show strong nonlinearity, separation, and cross-direction variability, which make flow field reconstruction from sparse sensors difficult. This study develops a learning-from-observation framework using wind-tunnel experimental data obtained by Particle Image Velocimetry (PIV) and compares Kriging interpolation with three deep learning models: UNet, Vision Transformer Autoencoder (ViTAE), and Conditional Wasserstein GAN (CWGAN). We evaluate two training strategies, single wind-direction training (SDT) and mixed wind-direction training (MDT), across sensor densities from 5 to 30, test robustness under sensor position perturbations of plus or minus 1 grid, and optimize sensor placement via Proper Orthogonal Decomposition with QR decomposition. Results show that deep learning methods can reconstruct rooftop wind fields from sparse sensor data effectively. Compared with Kriging interpolation, the deep learning models improved SSIM by up to 32.7%, FAC2 by 24.2%, and NMSE by 27.8%. Mixed wind-direction training further improved performance, with gains of up to 173.7% in SSIM, 16.7% in FAC2, and 98.3% in MG compared with single-direction training. The results also show that sensor configuration, optimization, and training strategy should be considered jointly for reliable deployment. QR-based optimization improved robustness by up to 27.8% under sensor perturbations, although with metric-dependent trade-offs. Training on experimental rather than simulated data also provides practical guidance for method selection and sensor placement in different scenarios.

State-of-the-art text-to-image diffusion models can produce impressive visuals but may memorize and reproduce training images, creating copyright and privacy risks. Existing prompt perturbations applied at inference time, such as random token insertion or embedding noise, may lower copying but often harm image-prompt alignment and overall fidelity. To address this, we introduce two complementary methods. First, Region-Aware Prompt Augmentation (RAPTA) uses an object detector to find salient regions and turn them into semantically grounded prompt variants, which are randomly sampled during training to increase diversity, while maintaining semantic alignment. Second, Attention-Driven Multimodal Copy Detection (ADMCD) aggregates local patch, global semantic, and texture cues with a lightweight transformer to produce a fused representation, and applies simple thresholded decision rules to detect copying without training with large annotated datasets. Experiments show that RAPTA reduces overfitting while maintaining high synthesis quality, and that ADMCD reliably detects copying, outperforming single-modal metrics.

What is a diffusion model actually doing when it turns noise into a photograph? We show that the deterministic DDIM reverse chain operates as a Partitioned Iterated Function System (PIFS) and that this framework serves as a unified design language for denoising diffusion model schedules, architectures, and training objectives. From the PIFS structure we derive three computable geometric quantities: a per-step contraction threshold $L^*_t$, a diagonal expansion function $f_t(\lambda)$ and a global expansion threshold $\lambda^{**}$. These quantities require no model evaluation and fully characterize the denoising dynamics. They structurally explain the two-regime behavior of diffusion models: global context assembly at high noise via diffuse cross-patch attention and fine-detail synthesis at low noise via patch-by-patch suppression release in strict variance order. Self-attention emerges as the natural primitive for PIFS contraction. The Kaplan-Yorke dimension of the PIFS attractor is determined analytically through a discrete Moran equation on the Lyapunov spectrum. Through the study of the fractal geometry of the PIFS, we derive three optimal design criteria and show that four prominent empirical design choices (the cosine schedule offset, resolution-dependent logSNR shift, Min-SNR loss weighting, and Align Your Steps sampling) each arise as approximate solutions to our explicit geometric optimization problems tuning theory into practice.

As virtual try-on (VTON) systems become increasingly important in fashion e-commerce, there is a growing need for reliable reference-free evaluation methods, since ground-truth images of the same person wearing the target garment are typically unavailable in real-world scenarios. To address this challenge, we propose VTON-IQA, a reference-free framework for human-aligned image quality assessment without requiring ground-truth images. To model human perceptual judgments, we construct VTON-QBench, a large-scale human-annotated benchmark comprising 62,688 try-on images generated by 14 representative VTON models and 431,800 quality annotations collected from 13,838 qualified annotators. To the best of our knowledge, this is the largest dataset to date for human subjective evaluation in VTON. Extensive experiments show that VTON-IQA achieves reliable human-aligned image quality assessment. Moreover, we conduct a comprehensive benchmark evaluation of 14 representative VTON models using VTON-IQA.

Continuous emotion recognition in terms of valence and arousal under in-the-wild (ITW) conditions remains a challenging problem due to large variations in appearance, head pose, illumination, occlusions, and subject-specific patterns of affective expression. We present a multimodal method for valence-arousal estimation ITW. Our method combines three complementary modalities: face, behavior, and audio. The face modality relies on GRADA-based frame-level embeddings and Transformer-based temporal regression. We use Qwen3-VL-4B-Instruct to extract behavior-relevant information from video segments, while Mamba is used to model temporal dynamics across segments. The audio modality relies on WavLM-Large with attention-statistics pooling and includes a cross-modal filtering stage to reduce the influence of unreliable or non-speech segments. To fuse modalities, we explore two fusion strategies: a Directed Cross-Modal Mixture-of-Experts Fusion Strategy that learns interactions between modalities with adaptive weighting, and a Reliability-Aware Audio-Visual Fusion Strategy that combines visual features at the frame-level while using audio as complementary context. The results are reported on the Aff-Wild2 dataset following the 10th Affective Behavior Analysis in-the-Wild (ABAW) challenge protocol. Experiments demonstrate that the proposed multimodal fusion strategy achieves a Concordance Correlation Coefficient (CCC) of 0.658 on the Aff-Wild2 development set.

Topology is critical in tubular structures such as blood vessels, nerve fibers, and road networks, where connectivity and loop structure govern downstream functional analysis. Vision-Language Models (VLMs) are promising candidates for understanding such structures, given their reasoning and grounding capabilities. To probe their topological perception, we systematically evaluate leading closed- and open-source VLMs on localizing and classifying four canonical topological anomalies (broken/spurious connections, missing/extra branches) in tubular-network segmentation masks. They perform nearly at random, indicating that topology-aware perception is largely absent from current general-purpose VLMs. As no existing resource pairs segmentation masks with localized anomaly annotations, we build an automated, multi-domain data-curation pipeline that synthesizes diverse topological perturbations with verifiable Betti-number annotations across graduated difficulty levels, yielding the first systematic benchmark with a large-scale training set and held-out in-distribution (ID) and out-of-distribution (OOD) test suites. Building on this benchmark, we introduce Topo-R1, centered on a topology-aware composite reward that jointly scores localization, classification, and skeleton-level structural fidelity. Supervised fine-tuning cold-starts schema-compliant outputs, and Group Relative Policy Optimization (GRPO) then optimizes the policy against this reward, steering predictions toward topologically meaningful structure rather than superficial pixel overlap. Extensive experiments show that Topo-R1 substantially outperforms general-purpose VLMs and matches or exceeds supervised baselines across ID, OOD, and real-segmentation-output protocols, establishing a strong foundation for VLM-based topological understanding of structured visual data.

Medical image segmentation (MIS) is a fundamental component of computer-assisted diagnosis and clinical decision support. Over the past decade, numerous architectures specifically tailored to medical imaging have emerged to address domain-specific challenges such as low contrast, small anatomical structures, and limited annotated data. In parallel, rapid progress in computer vision has produced highly capable general-purpose vision models (GP-VMs) originally designed for natural images. Despite their strong performance on standard vision benchmarks, their effectiveness for MIS remains insufficiently understood. In this controlled empirical study, we examine whether specialized medical segmentation architectures (SMAs) provide systematic advantages over modern GP-VMs for 2D MIS. We compare eleven SMAs and GP-VMs using a unified training and evaluation protocol across three heterogeneous datasets, each covering different 2D imaging modalities, class structures, and data characteristics. Beyond segmentation performance, we employ linear mixed-effects models (LMMs) for statistical assessment and qualitative Grad-CAM visualizations to investigate explainability (XAI)-related model behavior. Our results imply that, for the analyzed settings, GP-VMs achieve performance comparable to that of specialized MIS models. Moreover, XAI analyses indicate that GP-VMs are capable of identifying clinically relevant structures despite the absence of explicit domain-specific architectural priors. These findings suggest that GP-VMs can represent a viable alternative to domain-specific methods for 2D MIS, highlighting the importance of informed model selection. All code and resources are available at GitHub.