Papers

Text-to-image (T2I) diffusion models have the ability to build high-quality pictures from text prompts, but they pose safety concerns because they can generate offensive or disturbing imagery when provided with harmful inputs. Existing safety filters typically rely on text-based classifiers or image-based checkers that completely block the output upon detecting a threat, issuing an explicit allow/block feedback signal to the user. This binary strategy leaves models vulnerable to adversarial attacks that alter keywords to bypass detection, and it causes high false-alarm rates that degrade the experience for benign users. To address such vulnerabilities, we propose Disciplined Diffusion (DDiffusion), a novel robust text-to-image diffusion that counters Not Safe For Work (NSFW) generation by uncovering implicit malicious semantics in prompt embeddings. DDiffusion leverages a semantic retrieval mechanism to evaluate prompts against concept distributions rather than relying on brittle pairwise similarity. Furthermore, it employs a localization method during the diffusion process to selectively edit only the harmful regions of the generated image. By returning locally sanitized images instead of applying uniform blocking, DDiffusion suppresses malicious content while preserving generation fidelity for benign prompts and avoiding the binary allow-deny signal on which existing probing attacks rely.

Adversarial examples in machine learning are typically generated using gradients, obtained either directly through access to the model or approximated via queries to it. In this paper, we propose a much simpler approach to craft adversarial examples, drawing inspiration from insights of explainable machine learning. In particular, we design adversarial image filters that are based on classic edge detection algorithms but optimized to deceive learning models. The resulting untargeted attacks are transferable and require only a single pass over the input. Empirically, we find that 3x3 filters already enable success rates between 30% and 80% on different neural networks. Compared to related approaches using generative models for crafting adversarial examples, we reduce the number of parameters by five orders of magnitude, resulting in a very efficient attack. When investigating the parameters of the learned filters, we observe interesting properties such as a high transferability between models and structures common to classic image filters. Our results provide further insights into the vulnerability of neural networks and their fragility to malicious noise.

Mandibular reconstruction with vascularized bone grafts is complicated by donor-host nonunion, and virtual surgical planning produces a geometric plan rather than optimizing for bone-union propensity at the donor-host interface. We present OsteoOpt++, an image-to-decision planning loop for patient-specific mandibular reconstruction. Pre-operative computed tomography (CT) is converted into a personalized digital twin through template-to-patient registration and CT-derived updates of the muscle and temporomandibular-joint parameters. Bayesian optimization with an expected-improvement-plus acquisition rule then searches six clinically controllable cut-plane and donor-positioning variables under an apposition-driven objective and a safety-factor-regularized variant. The workflow was evaluated on three generic defects (body, symphysis, and ramus-body) and four patient-specific cases, three of which were used for optimization and all four for retrospective longitudinal spatial analysis. In the generic cases, against the surgeon's geometric plan, cycle-averaged donor-mandible apposition increased by up to 29 percentage points; in the patient-specific cases, against the surgeon-implemented day-5 postoperative configuration, by up to 26 percentage points. A +/-10% sensitivity analysis over eleven modeling parameters capped the change in the apposition-driven objective at approximately 3% (generic) and approximately 4% (patient-specific), and across the four longitudinal cases the Dice overlap between predicted apposition and year-1 bone formation ranged from 70.1% to 84.9%, with centroid shifts of 0.24 to 1.82 mm. Together, these results support the feasibility-stage use of OsteoOpt++ to compare candidate reconstructions using apposition-derived predictions of bone-union propensity. The optimization and patient-specific modeling code is open source at https://github.com/hamidreza-aftabi/OsteoOpt.

The performance of state-of-the-art object detectors degrades significantly under adverse weather, causing a safety-critical domain shift problem for autonomous vehicles. Recent efforts address this problem by relying on synthetic data to train the object detectors, which limits their real-world applicability. Meanwhile, pseudo-labeling is widely used for cross-dataset domain adaptation problems. However, these methods have not been exploited by weather-based domain adaptation approaches due to the noisy nature of such labels generated under harsh weather conditions. In this paper, we propose two new approaches to mitigate this weather-induced domain shift. First, we propose a Weather-Induced pseudo Label Denoising (WILD) framework that filters noisy pseudo labels generated by real data captured under adverse weather conditions. Second, we develop a novel hybrid training methodology, WILD SAM, that exploits both pseudo-label denoising and simulation-based training solutions while using real-data from the target harsh-weather domain. We validate both proposed approaches, WILD and WILD SAM, on the recently released Four Seasons dataset across rainy and snowy scenarios. Experiments show that the proposed frameworks improve Average Precision (AP) up to 13% and significantly reduce the weather-induced performance gap relative to the baseline. The code is available at: https://github.com/Kh-Hamed/WILD-SAM

Propagation-based X-ray phase-contrast imaging (PBI) enables high-contrast visualization of lung structures and holds strong medical potential. However, safe translation to the clinic will require a substantial radiation dose reduction, which inevitably increases image noise. Supervised convolutional-neural-network-based denoising can restore image quality but depends on paired low- and high-dose datasets, which are rarely available in practice. Self-supervised methods avoid this limitation, yet most are not well adapted to the inverse problem of PBI computed tomography (CT). We introduce Neighbor2Inverse, a self-supervised denoising framework designed for low-dose PBI-CT that generalizes to clinical CT. Building on the Neighbor2Neighbor principle, each noisy projection is subsampled into two variants that preserve structural information but contain independent noise realizations. These are reconstructed separately, and the resulting pairs are used to train a denoising network directly in the image domain. We benchmark the proposed method against established analytical and self-supervised denoising approaches. In region-of-interest PBI CT experiments, Neighbor2Inverse achieves superior noise suppression while preserving fine structural details, as demonstrated by improved contrast-to-noise ratio, spatial resolution, and composite image quality metrics. Competitive performance is also observed on clinical CT data under simulated low-dose conditions. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. Code, data, and interactive figures are available at https://github.com/J-3TO/Neighbor2Inverse.

Outlier Exposure (OE) is among the strongest training-based OOD detectors on standard benchmarks but exhibits scorer-dependent tradeoffs (e.g., strong on MSP, weak on KNN) and requires curated auxiliary data. We show why OE works: its features sit at the same geometric locus as real near-OOD data, with the boundary-adjacent quartile driving nearly all of OE's gain. OE is boundary calibration, not OOD coverage. GEODE (GEOmetry-preserving DEtection) replicates this calibration synthetically through an angle-adaptive norm loss in which targets scale per-sample with cosine similarity to the nearest class mean, preserving feature geometry where boundary structure matters. Four theorems grounded in neural collapse justify the design. GEODE works across all seven standard scorers on CIFAR-10 (near-OOD AUROC 89.0-92.3, far-OOD reaching 93.05; no catastrophic failure on any scorer). Since the OOD regime is unknown at deployment, this is the test that matters. GEODE outperforms vanilla CE at matched epoch counts. Combined with OE, GEODE reaches 95.0 MSP / 94.8 KNN on CIFAR-10 and beats OE on every scorer on CIFAR-100. The gains hold on WRN-28-10 (+4.5 Energy, 3 seeds). Unlike methods that push OOD into the classifier null space (e.g., PFS, 14.38 KNN AUROC, worse than random), GEODE's adaptive target preserves the geometry that distance-based scorers depend on.

This paper tackles the problem of physics-aware human motion synthesis in a dynamic scene. Unlike existing works which mainly tend to generate physically unrealistic motions due to limited contact modeling, typically restricted to hands, in this paper, we introduce a physics-aware human motion generation framework that explicitly models the full spectrum of human-related forces, including human-object, human-scene, and internal body dynamics.~Our method imposes soft physical constraints to maintain force and torque balance, ensuring physically grounded motion synthesis. We further propose a novel continuous distance-based force model that generalizes contact modeling to arbitrary surfaces, capturing interactions not only with static environments but also with dynamic, moving objects. Extensive experiments show that our approach significantly improves physical plausibility and generalizes well to complex scenes, setting a new benchmark for physically consistent human motion generation.

Multimodal Emotion Recognition (MER) is critical for interpreting real-world interactions. While Multimodal Large Language Models (MLLM) have shown promise in MER, their internal decision-making mechanisms under modality conflict and missingness remain largely underexplored. In this paper, to systematically investigate these behaviors, we introduce EmoMM, a comprehensive benchmark featuring modality-aligned, conflict, and missing subsets. Through extensive evaluation, we uncover a Video Contribution Collapse (VCC) phenomenon, where MLLM marginalize video evidence due to high token redundancy and modality preferences. To address this, we propose Conflict-aware Head-level Attention Steering (CHASE), a lightweight mechanism that detects modality conflicts and performs inference-time attention steering, effectively mitigating decision bias without retraining the backbone. Experimental results demonstrate that CHASE consistently improves performance across various settings, significantly enhancing the reliability of MLLM in complex affective scenarios.

Using multimodal foundation models to analyze table images is a high-value yet challenging application in consumer and enterprise scenarios. Despite its importance, current evaluations rely largely on structured-text tables or clean rendered images, leaving the visual complexity of in-the-wild table images underexplored. Such images feature varied layouts and diverse domains that demand sophisticated structural perception and numerical reasoning. To bridge this gap, we introduce WildTableBench, the first question-answering benchmark for naturally occurring table images from real-world settings. WildTableBench comprises 402 high-information-density table images collected from online forums and websites across diverse domains, together with 928 manually annotated and verified questions spanning 17 subtypes across five categories. We evaluate 21 frontier proprietary and open-source multimodal foundation models on this benchmark. Only one model exceeds 50% accuracy, while all remaining models range from 4.1% to 49.9%. We further conduct diagnostic analyses to characterize model failures and reveal persistent weaknesses in structural perception and reasoning. These results and analyses provide useful insights into current model capabilities and establish WildTableBench as a valuable diagnostic benchmark for table image understanding. Dataset: https://huggingface.co/datasets/jzhuang/WildTableBench Code: https://github.com/hjzhe/WildTableBench Leaderboard: https://hjzhe.github.io/WildTableBench

The record of the beginning of the most widespread legal system in the world is contained in millions of pages of handwritten text. Most of the records of the first centuries of the Anglo-American legal system are hand-written in a highly abbreviated form of medieval Latin which only a few dozen scholars in the world are trained to read. In this interdisciplinary project, we construct a dataset of 4029 lines of text across 193 medieval criminal and civil cases. We then use the dataset to train an open-source end-to-end pipeline for transcribing these manuscripts. We first train standard neural network architectures for line segmentation and handwriting recognition (R-Blla and CNN+LSTM with CTC decoding, respectively) and show that they can already achieve 79% word accuracy, despite the relatively small training set and the challenge of expanding abbreviations. We then demonstrate that simple post-processing significantly boosts accuracy: adding an n-gram language model to the CTC decoder improves word accuracy to 82%, while asking Gemini Pro 3 to correct mistakes boosts accuracy to 88%. Finally, we compare the CNN+LSTM architecture with TrOCR, a transformer-based OCR architecture, demonstrating that TrOCR shows comparable word accuracy but worse character accuracy due to its over-willingness to guess, making it harder for humans to infer the correct reading. We incorporated our pipeline into a web portal (glyphmachina.com), opening up the English legal tradition to legal scholars, medievalists, and students.

Earth system science is producing increasingly large, high-dimensional datasets from both physics-based and AI-driven models. While embedding-based representations make these data searchable and serve as foundational building blocks for AI-driven discovery engines, nearest neighbors in latent spaces are not automatically scientifically meaningful. They may reflect real meteorological structures, or simply artifacts of preprocessing, geography, or model bias. Researchers therefore need visual tools to inspect latent space organization, trace search results back to physical evidence, and evaluate candidate representations against one another. We present an open source visual analytics workbench designed to support this provenance-aware scientific retrieval workflow. The system links distinct embedding experiments to shared source data, metadata, spatial contexts, and model configurations. It enables interactive retrieval strategy design by allowing users to issue image-level and localized patch-level queries, apply multi-constraint filters, and inspect analogs through familiar meteorological views. This facilitates a discovery loop where scientists characterize a phenomenon in a well-understood dataset and use its latent signature to probe larger archives. While we demonstrate the workbench through a tropical cyclone retrieval scenario using a vision foundation model (DINOv3) on ERA5 data, the framework is model-agnostic and designed to integrate with other embedding architectures in the future. Finally, we evaluate its out-of-core retrieval backend, demonstrating that interactive visual search over tens of millions of embeddings is highly scalable on commodity hardware.

Background: Sensitivity of AI-assisted lung nodule detection systems is known to vary with CT acquisition parameters including radiation dose, reconstruction kernel, and slice thickness. However, the dependence of detection probability on nodule position within the reconstruction cycle – the z-phase – has not, to the author's knowledge, been characterized for deep learning-based detection systems. Methods: A retrospective analysis was performed using the LIDC-IDRI dataset. Detection results from a previously validated 154-case perturbation study were re-analyzed. For each consensus nodule (>=4-reader agreement), z-phase was defined as the fractional position of the nodule center within the reconstruction cycle, folded to [0, 0.5]. Detection sensitivity was stratified by z-phase bin, reconstruction interval (1mm, 3mm, 5mm), and by the ratio of reconstruction interval to nodule diameter (d/D). Results: At 5mm reconstruction interval, sensitivity was 71.6% vs 84.8% at 1mm baseline. Within the 5mm condition, sensitivity varied by 17.6 percentage points across z-phase bins. Stratified by d/D ratio, sensitivity was 92.4% for d/D < 0.5, 78.0% for 0.5 <= d/D < 1.0, and 61.4% for d/D >= 1.0, with a systematic z-phase effect present only in the d/D >= 1.0 stratum. Conclusions: AI detection sensitivity depends on the ratio of reconstruction interval to nodule diameter. When this ratio approaches or exceeds 1.0 – as occurs for 3-6mm nodules at 5mm reconstruction – z-phase becomes the dominant source of per-study detection variance. This stochastic effect is invisible to protocol-level quality metrics and not reflected in AI confidence scores.

We introduce energy-based constraint networks – a modality-agnostic architecture that learns structural coherence from contrastive pairs. The system processes frozen encoder embeddings through a state-space model with dual-head attention, producing a scalar energy measuring structural consistency alongside per-position energy scores that localize violations. Multiple independently trained branches detect different violation types and compose at inference without interference. We demonstrate the framework in two domains. In text, the system achieves 93.4% accuracy on trained corruption types and 87.2% on 9 unseen types, using frozen BERT and 7.4M trainable parameters. In vision, the same architecture achieves competitive deepfake detection: 0.959 AUC on FaceForensics++ Deepfakes and 0.870 on Celeb-DF without any Celeb-DF training data, using frozen DINOv2 and 3.6M parameters per branch. The framework supports flexible training: branches learn from designer-specified corruptions, real-world paired data, or both. Composable branches require representation compatibility – a finding validated through extensive experimentation where five incompatible approaches failed before the compatible one succeeded. The architecture is encoder-agnostic and domain-agnostic: changing the domain requires only new corruption strategies; changing the encoder requires only a new input projection layer. To our knowledge, this is the first architecture to learn within-modality structural coherence as an explicit energy landscape with per-position decomposition, and to demonstrate that the same architecture transfers across modalities via corruption respecification alone.

Flow matching provides a highly effective framework for generative modeling, yet estimating the uncertainty of its generated samples remains a fundamental challenge. Existing methods rely on auxiliary variance heads, model ensembles, or iterative covariance propagation, which invariably introduce accumulated errors and structural interference along the noise-to-image trajectory. We resolve these limitations by adapting Tweedie's formula to the linear flow matching interpolant, yielding an exact, closed-form identity for the posterior covariance. Because this identity relies exclusively on the divergence of the learned velocity field, it can be evaluated post hoc on pretrained models without any architectural modifications. This analytical formulation proves exceptionally powerful for single-step generative frameworks. By computing the end-to-end posterior covariance in one forward pass, our approach completely bypasses the compounding errors inherent to sequential integration. This mechanism fundamentally enhances robustness, providing highly accurate and stable uncertainty estimates while preventing the numerical degradation typical of multi-step approximations. We validate our framework across benchmark datasets, including CIFAR-10 and real brain MRI scans. The resulting divergence-based uncertainty maps are highly interpretable and tightly correlated with empirical reconstruction errors. Crucially, in medical imaging, this precise localization of structural ambiguity provides essential decision support for high-stakes clinical tasks such as tumor boundary delineation and neurosurgical planning.

Diffusion models have become the foundation of modern generative systems, with most research focusing primarily on improving generation efficiency and output quality. The timestep embedding component is a crucial part of the diffusion pipeline, which provides a temporal conditioning signal to the denoising network, enabling it to adapt its predictions across different noise levels throughout the process. Despite their potential to contain substantial information, timestep embeddings remain underexplored in current research, especially for security risks and reliable provenance. To fill this gap, we introduce Shadow Timestep Embedding (STE), a novel mechanism that investigates the underutilized temporal space for malicious information injection into diffusion models. In particular, when zooming in on the timestep embedding space, we find that different timesteps exhibit distinct representational capabilities that can encode side-channel information. Moreover, such encoded information can be utilized for attack and defense purposes through the scheduler interface. We present a theoretical analysis of timestep embeddings as position-encoding mappings and derive a mutual coherence evaluation that explains the separability of disjoint timestep intervals. Our findings reveal the diffusion model's timestep as a powerful side channel for carrying dedicated information, motivating new directions for adversarial generative modeling by understanding the temporal dimension.

Coherent Point Drift (CPD) is a representative probabilistic framework for unsupervised non-rigid point set registration. Its standard non-rigid M-step, however, relies on a point-indexed Gaussian-kernel system whose size grows with the number of moving points, making deformation estimation computationally heavy for large point sets and difficult to control in complexity during registration. To address these limitations, we propose Analytic-CPD, a new unsupervised non-rigid registration framework that gives CPD a structured analytic reformulation. Analytic-CPD preserves the CPD posterior correspondence layer, but lifts the M-step from point-indexed kernel displacement estimation to structured analytic mapping estimation. By coupling the Gaussian-mixture posterior mechanism of CPD with Structured Analytic Mappings (SAM), the method obtains a deformation model whose coefficient dimension is governed by the ambient dimension and analytic order rather than by the number of moving points. More importantly, deformation estimation is organized over an interpretable hierarchy of analytic function spaces, so the analytic order can be increased progressively as posterior correspondences become more reliable. We implement this idea through an increasing-degree continuation strategy with decreasing stage lengths: low-order analytic maps first stabilize the posterior correspondence structure, while higher-order modes later refine nonlinear residual deformation. Experiments on controlled model-matched, smooth model-mismatch, and registered human-shape data demonstrate the effectiveness and favorable accuracy–efficiency performance of Analytic-CPD.

Flow matching (FM) trains a time-dependent vector field that transports samples from a simple prior to a complex data distribution. However, for high-dimensional images, each training sample supervises only a single trajectory and intermediate point, yielding an extremely sparse and high-variance training signal. This under-constrained supervision can cause flow collapse, where the learned dynamics memorize specific source-target pairings, mapping diverse inputs to overly similar outputs, failing to generalize. We introduce Posterior-Augmented Flow Matching (PAFM), a theoretically grounded generalization of FM that replaces single-target supervision with an expectation over an approximate posterior of valid target completions for a given intermediate state and condition. PAFM factorizes this intractable posterior into (i) the likelihood of the intermediate under a hypothesized endpoint and (ii) the prior probability of that endpoint under the condition, and uses an importance sampling scheme to construct a mixture over multiple candidate targets. We prove that PAFM yields an unbiased estimator of the original FM objective while substantially reducing gradient variance during training by aggregating information from many plausible continuation trajectories per intermediate. Finally, we show that PAFM improves over FM by up to 3.4 FID50K across different model scales (SiT-B/2 and SiT-XL/2), different architectures (SiT and MMDiT), and in both class and text conditioned benchmarks (ImageNet and CC12M), with a negligible increase in the compute overhead. Code: https://github.com/gstoica27/PAFM.git.

While autoregressive Large Vision-Language Models (LVLMs) demonstrate remarkable proficiency in multimodal tasks, they face a "Visual Signal Dilution" phenomenon, where the accumulation of textual history expands the attention partition function, causing visual attention to decay inversely with generated sequence length. To counteract this, we propose Persistent Visual Memory (PVM), a lightweight learnable module designed to strengthen sustained, on-demand access to visual evidence. Integrated as a parallel branch alongside the Feed-Forward Network (FFN) in LVLMs, PVM establishes a distance-agnostic retrieval pathway that directly provides visual embeddings for enhanced visual perception, thereby structurally mitigating the signal suppression inherent to deep generation. Extensive experiments on Qwen3-VL models demonstrate that PVM brings notable improvements with negligible parameter overhead, delivering consistent average accuracy gains across both 4B and 8B scales, particularly in complex reasoning tasks that demand persistent visual perception. Furthermore, in-depth analysis reveals that PVM shows improved robustness in longer generations and accelerates internal prediction convergence.

In this paper, we present Generative Language-Image Pre-training (GenLIP), a minimalist generative pretraining framework for Vision Transformers (ViTs) designed for multimodal large language models (MLLMs). To better align vision encoders with the autoregressive nature of LLMs, GenLIP trains a ViT to predict language tokens directly from visual tokens using a standard language modeling objective, without contrastive batch construction or an additional text decoder. This design offers three key advantages: (1) Simplicity: a single transformer jointly models visual and textual tokens; (2) Scalability: it scales effectively with both data and model size; and (3) Performance: it achieves competitive or superior results across diverse multimodal benchmarks. Trained on 8B samples from Recap-DataComp-1B, GenLIP matches or surpasses strong baselines despite using substantially less pretraining data. After continued pretraining on multi-resolution images at native aspect ratios, GenLIP further improves on detail-sensitive tasks such as OCR and chart understanding, making it a strong foundation for vision encoders in MLLMs.

Gaze estimation methods commonly use facial appearances to predict the direction of a person gaze. However, previous studies show three major challenges with convolutional neural network (CNN)-based, transformer-based, and contrastive language-image pre-training (CLIP)-based methods, including late fusion of image features, lack of factor-aware conditioning, and impractical capacity scaling. To address these challenges, we propose Globally-conditioned Multi-scale Gaze estimation (GMGaze), which leverages a multi-scale transformer architecture. Specifically, the model first introduces semantic prototype conditioning, which modulates the CLIP global image embedding using four learned prototype banks (i.e., illumination, background, head pose and appearance) to generate two complementary context-biased global tokens. These tokens, along with the CLIP patch and CNN tokens, are fused at the first layer. This early unified fusion prevents information loss common in late-stage merging. Finally, each token passes through sparse Mixture-of-Experts modules, providing conditional computational capacity without uniformly increasing dense parameters. For cross-domain adaptation, we incorporate an adversarial domain adaptation technique with a feature separation loss that encourages the two global tokens to remain de-correlated. Experiments using four public benchmarks (MPIIFaceGaze, EYEDIAP, Gaze360, and ETH-XGaze) show that GMGaze achieves mean angular errors of 2.49$^\circ$, 3.22$^\circ$, 10.16$^\circ$, and 1.44$^\circ$, respectively, outperforming previous baselines in all within-domain settings. In cross-domain evaluations, it provides state-of-the-art (SOTA) results on two standard transfer routes.

With the development of deep learning, medical image processing has been widely used to assist clinical research. This paper focuses on the denoising problem of low-dose computed tomography using deep learning. Although low-dose computed tomography reduces radiation exposure to patients, it also introduces more noise, which may interfere with visual interpretation by physicians and affect diagnostic results. To address this problem, inspired by Cycle-GAN for unsupervised learning, this paper proposes an end-to-end unsupervised low-dose computed tomography denoising framework. The proposed framework combines a U-Net structure for multi-scale feature extraction, an attention mechanism for feature fusion, and a residual network for feature transformation. It also introduces perceptual loss to improve the network for the characteristics of medical images. In addition, we construct a real low-dose computed tomography dataset and design a large number of comparative experiments to validate the proposed method, using both image-based evaluation metrics and medical evaluation criteria. Compared with classical methods, the main advantage of this paper is that it addresses the limitation that real clinical data cannot be directly used for supervised learning, while still achieving excellent performance. The experimental results are also professionally evaluated by imaging physicians and meet clinical needs.

Key-Value (KV) cache has become a de facto component of modern Large Vision-Language Models (LVLMs) for inference. While it enhances decoding efficiency in Large Language Models (LLMs), its direct adoption in LVLMs introduces substantial GPU memory overhead due to the large number of vision tokens processed during the prefill stage. To tackle this problem, we propose LightKV, a novel approach that reduces KV cache size by exploiting the redundancy among vision-token embeddings. Guided by text prompts, LightKV employs cross-modality message passing to aggregate informative messages across vision tokens and progressively compress them during prefill. This prompt-aware guidance distinguishes our method from prior vision-only compression strategies. We evaluate LightKV on eight open-source LVLMs across eight public benchmark datasets, e.g., MME and SeedBench. Experimental results demonstrate that with only 55% of the original vision tokens, LightKV (a) halves the vision-token KV cache size, (b) reduces computation by up to 40%, and (c) preserves general-purpose performance while significantly outperforming existing baselines.

3D world generation is essential for applications such as immersive content creation or autonomous driving simulation. Recent advances in 3D world generation have shown promising results; however, these methods are constrained by grid layouts and suffer from inconsistencies in object scale throughout the entire world. In this work, we introduce a novel framework, Map2World, that first enables 3D world generation conditioned on user-defined segment maps of arbitrary shapes and scales, ensuring global-scale consistency and flexibility across expansive environments. To further enhance the quality, we propose a detail enhancer network that generates fine details of the world. The detail enhancer enables the addition of fine-grained details without compromising overall scene coherence by incorporating global structure information. We design the entire pipeline to leverage strong priors from asset generators, achieving robust generalization across diverse domains, even under limited training data for scene generation. Extensive experiments demonstrate that our method significantly outperforms existing approaches in user-controllability, scale consistency, and content coherence, enabling users to generate 3D worlds under more complex conditions.

Urban perception describes how people subjectively evaluate urban environments, shaping how cities are experienced and understood. Existing computational approaches primarily model urban perception directly from street view images, but largely ignore the human perceptual process through which such judgments are formed. In this paper, we introduce Place Pulse-Gaze, an urban perception dataset that augments street view images with synchronized eye-tracking recordings and individual perception labels. Based on this dataset, we propose a Gaze-Guided Urban Perception Framework to study how gaze behavior contributes to the modeling of subjective urban perception. The framework systematically investigates three complementary settings: gaze-only modeling, gaze fusion with explicit semantic scene representations, and gaze fusion with implicit richer visual representations. Experiments show that gaze alone already carries useful predictive signals for subjective urban perception, and that integrating gaze with scene representations further improves prediction under both semantic and richer visual representations. Overall, our findings highlight the importance of incorporating human perceptual processes into urban scene understanding and open a direction for gaze-guided multimodal urban computing.

Edge detection refers to identifying points in a digital image where intensity changes sharply, indicating object boundaries or structural features. Corners are locations where gray-level intensity changes abruptly in multiple directions and are widely used in feature extraction, object tracking, and 3D modeling. In this study, we present a quantum implementation of Sobel-based edge detection and Harris-style corner detection. Two quantum image encoding methods - Flexible Representation of Quantum Images (FRQI) and Quantum Probability Image Encoding (QPIE) - are used to encode the input data and are comparatively analyzed. The proposed approach introduces a quantum gradient computation scheme based on lag-2 differences, enabling the evaluation of gradient-like features in superposition. To improve detection quality and reduce false positives, a classical post-processing step is applied to candidate corner points identified by the quantum circuit. Results show that the proposed quantum circuits produce outputs consistent with classical Sobel and Harris operators. Furthermore, the QPIE-based configuration yields more stable and coherent results than FRQI, especially under limited measurement shots. While gradient computation can be performed efficiently at the circuit level, the overall cost remains dominated by state preparation, measurement, and classical post-processing. All experiments are conducted under noiseless simulation, and performance on NISQ hardware may be affected by noise and measurement limitations. Therefore, this work demonstrates a functional and scalable quantum realization of classical edge and corner detection methods rather than an end-to-end speedup.

Single-point supervised infrared small target detection (IRSTD) drastically reduces dense annotation costs. Current state-of-the-art (SOTA) methods achieve high precision by recovering mask supervision through explicit, offline pseudo-label construction, such as multi-stage active learning and physics-driven mask generation. In this paper, we study a minimalist alternative: generating point-to-mask supervision online through in-batch, point-anchored feature-affinity propagation. We instantiate this paradigm as GSACP, an end-to-end testbed that directly supervises the detector using hard-margin feature affinity gated by local image priors, entirely eliminating external label-evolution loops. This compact design, however, exposes an optimization bottleneck. Because the affinity target is generated from the same feature representation being optimized, training forms a self-referential loop. We theoretically formalize this as Self-Referential Propagation Drift, a representation-supervision entanglement that can sharpen true boundaries or distort the feature space to satisfy its own targets. To systematically isolate these failure modes, we apply a protocolized single-variable ablation procedure spanning local EMA teacher decoupling, hard-background contrastive separation, and adaptive support geometry. On the SIRST3 dataset, GSACP-Final establishes a new ultra-low false-alarm operating regime, achieving a highly competitive $0.6674$ mIoU while demonstrating a $38\% relative reduction in false-positive artifacts ($\mathrm{Fa}$) compared with PAL. By systematically deconstructing the end-to-end paradigm, we map its performance boundaries and show that in-batch feature propagation provides a compact alternative for deployment scenarios where false-alarm suppression is paramount.

Unsupervised deraining has attracted attention for its ability to learn the real-world distribution of rain without paired supervision. However, the lack of strong constraints makes it difficult for the network to converge, especially with the complex diversity of rain degradation. A key motivation is that high-quality deraining results occasionally emerge during training, which can be leveraged to guide the optimization process. To overcome these challenges, we introduce RGSUD (Reward-Guided Self-Reinforcement Unsupervised Image Deraining), comprising two key stages: reward recycling and self-reinforcement (SR) training. For the former stage, we propose an Image Quality Assessment (IQA)-based dynamic reward recycling mechanism that selects optimal derained outputs during training and continuously collects high-quality deraining images. In latter stage, we incorporate these rewards into the model's optimization process, constraining the optimization space and improving alignment between derained outputs and clean images. By leveraging IQA-based self-reinforced loss and dynamically updated rewards, we enhance the quality of synthesized pseudo-paired data and stabilize the optimization. Extensive experiments demonstrate that our method achieves SOTA performance across multiple datasets, including paired synthetic, paired real, and unpaired real images, outperforming existing unsupervised deraining approaches in both subjective and objective IQA metrics. Additionally, we show that the self-reinforcement strategy is adaptable to other unsupervised deraining methods and our deraining framework demonstrates strong generalization across existing supervised deraining networks.

Knee osteoarthritis (OA) assessment contains a natural label hierarchy between binary disease status and Kellgren–Lawrence (KL) severity. We study whether supervision at these two granularities changes learned 3D MRI representations. A shared encoder with OA and KL prediction heads is evaluated under single-OA, single-KL, and dual-head training across ResNet3D, M3T, and nnMamba backbones. Evaluation combines predictive metrics with paired statistical comparisons under Benjamini–Hochberg FDR control, latent severity-axis geometry, and saliency–cartilage overlap. Dual supervision yields significant KL-grading gains for ResNet3D and a significant OA AUC gain for M3T, whereas nnMamba retains stronger single-task performance. Representation analysis further shows an architecture-dependent effect: Dual strengthens label-aligned latent organization for ResNet3D and M3T, while nnMamba retains stronger alignment under single-task supervision. For the responsive backbones, Dual also produces descriptively higher saliency overlap with cartilage. These results show that coarse-to-fine supervision can reshape disease representations under noisy hierarchical labels, with benefits that depend on backbone architecture. Code is available at https://github.com/jukieCheung/coarse2fine-oa-mri.

Image editing instructions are heterogeneous: a color swap, an object insertion, and a physical-action edit all demand different spatial coverage and different reasoning depth, yet existing reasoning-based editors apply a single fixed inference recipe to every instruction. We argue that adaptivity along both the spatial and temporal axes is the missing degree of freedom, and we present PhysEdit, an editing framework built around this principle. PhysEdit introduces two inference-time modules that compose without retraining the backbone. At its core, (1) Complexity-Adaptive Reasoning Depth (CARD) predicts edit complexity directly from the instruction and reference image and allocates the reasoning step count N_r and reasoning-token length r per sample – turning a previously fixed inference schedule into a conditional-computation problem. CARD is supported by (2) a Spatial Reasoning Mask (SRM) that extracts an instruction-conditioned spatial prior from cross-attention to confine reasoning to regions that semantically require it. On the full 737-case ImgEdit Basic-Edit Suite, PhysEdit delivers a 1.18x wall-clock speedup (64.3s vs. 76.1s per sample) over a strong reasoning baseline while slightly improving instruction adherence (CLIP-T 0.2283 vs. 0.2266, +0.7%) and matching identity preservation within noise (CLIP-I 0.8246 vs. 0.8280). The speedup is category-dependent and reaches 1.52x on appearance-level edits, validating CARD's adaptive allocation as the principal source of efficiency gain. A 30-sample pilot with full ablations isolates the contribution of each module.

Temporal Video Grounding (TVG) aims to localize temporal moments in an untrimmed video that semantically correspond to given natural language queries. Recently, Graph Convolutional Networks (GCN) have been widely adopted in TVG to model temporal relations among video clips and enhance contextual reasoning by constructing clip-level graphs. Despite their effectiveness, existing GCN-based TVG methods encounter three critical bottlenecks: 1) Most methods construct graph nodes using either static or dynamic features alone, resulting in incomplete visual representation and overlooking complementary semantics, 2) Most methods construct temporal graphs in a query-agnostic manner, leading to inefficient feature interaction within the temporal graph representation, and 3) Most methods often suffer from a single-granularity semantic matching, while direct training on complex temporal localization task may lead to slow convergence and suboptimal precision. To address these challenges, we propose Static and Dynamic Graph Alignment Network (SDGAN). First, SDGAN jointly exploits static and dynamic visual features to construct two complementary temporal graphs and performs Position-wise Nodes Alignment, enabling more expressive and robust visual representation. Second, SDGAN introduces Query-Clip Contrastive Learning and Adaptive Graph Modeling to explicitly align visual clips with their corresponding textual queries, yielding query-aware visual representations. Third, SDGAN incorporates multi-granularity temporal proposals within Progressive Easy-to-Hard Training Strategy, effectively bridging coarse-grained semantic localization and fine-grained temporal boundary refinement. Extensive experiments on three benchmark datasets demonstrate that SDGAN achieves superior performance across complex TVG scenarios. Codes and datasets are available at https://github.com/ZhanJieHu/SDGAN.