Publication Radar.

Measurements of tree diameter at breast height (DBH) are fundamental to forest inventories, but manual surveys across entire stands or regions are impractical. UAV-borne laser scanning (ULS) enables scalable automated forest surveys, yet canopy occlusion makes accurate DBH estimation from ULS point clouds non-trivial. Limited stem-level point density of ULS data constrains approaches relying on detailed stem reconstruction, such as Quantitative Structure Models (QSMs) and circle fitting. Empirical machine learning methods based on predefined point cloud metrics are an alternative. New deep learning approaches offer an even more principled solution, allowing models to learn directly from point clouds. Sparse tensor networks have shown particular promise, transforming forestry tasks such as tree species classification. We investigate whether these strengths can be extended to estimate individual tree DBH from ULS data across four mixed-species temperate forest plots in Germany. We further advance existing sparse tensor networks by introducing TreeNext, a lightweight deep learning architecture for direct individual tree parameter estimation from voxelized LiDAR point clouds. We present a fully automated forest inventory pipeline from individual tree instance segmentation to DBH regression, and quantify the impact of segmentation errors by comparing performance on automatically segmented and ground truth instances. TreeNext achieves an $R^2$ of 0.8355 on automatically segmented trees and 0.9380 on ground truth segments, outperforming strong tabular baselines including Random Forest and XGBoost. Segmentation quality emerges as the dominant error source in the automated pipeline. TreeNext proves highly data-efficient, performing well even in small data scenarios. Our findings establish sparse tensor networks as a practical foundation for automated individual tree DBH estimation from ULS.

Humans navigate their environment based on their vision system, anticipating how the ground will respond before committing to a step. Self-supervised learning could provide legged robots the same ability, but the supervisory signal is typically chosen on intuition, and its effect on what the model learns is rarely examined. We present Terrain-Aware COntact Supervision (TACOS), which proposes a meaningful and consistent footstep instability score, based on the residual of a factor graph in legged state estimation. It projects the continuous score into preceding images to supervise a dense visual predictor. We propose two variants, a hard score that exploits hindsight and a causal soft score available immediately. Both metrics are compared against state-of-the-art alternatives in two independent evaluations that share no data or labels. On 1716 manually annotated contacts, both variants outperform every evaluated alternative at detecting instability and ranking its severity, improving pooled severity ranking by 28 % over the strongest baseline. On three outdoor test runs, our metric also trains the most accurate visual predictor, improving rank correlation by 58 % over the strongest alternative, and by a further 25 % when supplemented with onboard context. In a real-world experiment over a 50 m forest trajectory, the trained predictor handles non-geometric terrain properties such as mud and vegetation, while stepping around geometric obstacles like fallen branches. Code is available for reproducibility at https://github.com/norlab-ulaval/TACOS.

Robots, autonomous vehicles, and consumer 360° cameras use fisheye lenses for wide fields of view (FoVs). Simulation-ready digital twins require photorealistic rendering across the full FoV and an aligned collision mesh. However, prior fisheye Gaussian Splatting methods only rasterize up to 180° and do not produce meshes, while Gaussian surface reconstruction methods assume pinhole training images. We present GOFish, the first 3D Gaussian Splatting surface reconstruction method to train directly on fisheye images, rasterize FoVs beyond 180°, and extract an aligned mesh from the same scene. Calibrated densification recovers the surface across the entire FoV, while Calibrated Mip Filtering and Wide-Angle Appearance Modeling keep pixel blur and lens falloff out of the scene and mesh. Across the 200° FIORD and 205° FullCircle benchmarks, GOFish improves PSNR over the best prior method by 2.28 and 1.97 dB, respectively. It also outperforms fisheye rendering baselines within their own FoVs and achieves the highest mesh F-score on FIORD and the main ScanNet++ set. Together, the Gaussian scene and mesh form a LiDAR-free, simulation-ready digital twin that renders in real time, ray traces the full FoV, and provides collision geometry.

Event cameras, spiking cochleas and similar sensors emit asynchronous events, yet most learning systems first bin them into frames or packets, so the output depends on how the stream was sliced. Networks whose state is updated exactly at every event and evolves in closed form between events are invariant to this slicing. We ask what more makes the output (i) independent of the processing order of simultaneous events and (ii) continuous in the time stamps, even when two events cross in time, while (iii) staying sensitive to temporal order. For arbitrary nonlinear event models we prove that (i) and (ii) are each equivalent to pairwise commutation of the event updates, and that once updates commute, all order dependence comes from their non-commutation with the inter-event dynamics and vanishes linearly in the time lag. This identifies a class of time-mediated non-commutative (TM) models and yields a resolution-robustness bound for any model. We construct the minimal TM layer, a learnable, provably Lipschitz generalization of the Hassenstein-Reichardt correlator; prove that no single additive layer can uniformly represent ordered pair interactions, whereas a three-dimensional TM state can; show that each TM layer adds two levels of temporal nesting versus one for additive layers; and give a single-layer higher-order construction (TM-k) and a stacking rule preserving all three properties. We instantiate these results as the Clock-Free Event (CFE) layer, linear in the number of events, give a topology-free variant for temporal point processes with closed-form likelihoods, and specify a preregistered clock-robustness protocol with a physical-time threat model for spike-retiming attacks. In preliminary development-set experiments, the TM state beats an otherwise identical additive state by 3.5-9.1 points on DVS128 Gesture and by 24.5 and 30.6 points on SHD and Speech Commands.

Computing optimal transport maps between medical images plays a fundamental role in morphological analysis. However, existing methods are either difficult to scale to large 3D medical images or rely on regularized formulations that provide only approximate transport maps. To improve the accuracy of downstream medical image analysis, we propose a novel approach for estimating optimal transport maps by parameterizing the convex potential function. By construction, the resulting transport map is guaranteed to be curl-free, consistent with Brenier's theorem. We evaluate the proposed method on both 2D and 3D medical imaging datasets. Experimental results demonstrate that our approach achieves lower transport estimation errors and produces higher-quality image reconstructions than existing methods.

Two-dimensional (2D) intraglottal particle image velocimetry (PIV) of ex vivo phonation experiments quantifies vocal fold medial surface dynamics and the intraglottal flow field. However, it relies on manual segmentation because residual glare, low tissue-to-flow contrast, and tracer-particle speckle obscure the vocal fold boundaries. This study assessed how deep learning architecture and ImageNet pretraining affect automated vocal fold segmentation in these high-speed images. Four established architectures from three categories were benchmarked: the convolutional neural network (CNN)-based UNet and UNet++, the hybrid CNN-transformer-based TransUNet, and the pure transformer-based Swin-UNet, the latter two trained from scratch and from ImageNet-pretrained weights. Using 16,344 expert-annotated images from 227 phonation trials of excised canine larynges, the six pipelines were trained with five seeds under a standardized protocol and compared on 50 held-out test trials using trial-level Wilcoxon signed-rank tests with Holm correction. Mean Dice similarity coefficient (DSC) spanned only 0.9734 to 0.9764, so the pipelines differed mainly in boundary fidelity. Pretraining significantly improved the boundary F1 score (BF1) and 95th percentile Hausdorff distance (HD95) of the TransUNet and Swin-UNet, about four times more for the Swin-UNet. Once pretrained, the two performed similarly, although the Swin-UNet has only about a quarter of the trainable parameters. Without pretraining, the TransUNet had a significantly higher BF1 and a lower HD95 than both CNN-based pipelines and about half their share of images with HD95 > 50 pixels. To our knowledge, this is the first study to apply deep learning to vocal fold segmentation in 2D intraglottal PIV, and it offers a reference framework for segmenting experimental flow-visualization images with noisy optical conditions or highly deformable objects.

Food similarity underlies retrieval, recommendation, substitution, and other food computing tasks, yet food representations are typically evaluated by identity preservation. We ask what happens to food similarity beyond identity: how different visual representations organize foods into relational neighborhoods, what culinary relationships these neighborhoods reveal or obscure, and how consistently these structures emerge across representations. Indian cuisine provides a challenging setting in which visually similar foods may have different culinary identities, while foods within a culinary family can vary substantially in appearance. Using an expanded Khana dataset and seven pretrained visual representations spanning convolutional, transformer-based, self-supervised, and image-text aligned models, we evaluate image-to-image retrieval and image-to-dish neighborhoods, examining culinary-family structure, contrast cases, and cross-representation agreement. We also aggregate image representations to compare dish-level visual similarity across models and against recipe-derived similarity for the same dishes. Image-level neighborhoods vary substantially across representations, including similarities across culinary-family boundaries. At the dish level, aggregation reveals common visual structure despite local variation, but this structure corresponds less strongly with recipe-derived similarity. These findings show that recognition performance alone cannot characterize relational structure: food similarity depends on representation, modality, and abstraction level rather than being a fixed property of the food collection. Understanding these structures is therefore a prerequisite to combining multiple food information modalities into a coherent notion of similarity.

Deepfake detectors are almost always deployed as the single checkpoint that scored best on a validation split, yet the reliability of that choice is rarely measured. We measure it with PairCLIP-SWA, a single-backbone detector that adapts an OpenCLIP ViT-L/14 visual tower by progressive block unfreezing on a class-balanced corpus built by CLIP-similarity-guided one-to-one subsampling, and averages the fine-tuning trajectory in weight space. Trained in-domain on the IEEE SP Cup 2025 (DF-Wild-Cup) benchmark under the same protocol as the strongest published three-backbone ensemble, the weight-averaged detector reaches 97.04% accuracy, 98.64% AU-ROC and a 4.07% equal error rate, against 94.46%, 97.60% and 5.00%. Its training log shows why averaging matters: over the averaged epochs the non-averaged checkpoint swings across 5.34 points of validation accuracy while its validation loss more than doubles, whereas the average is never worse and is up to 5.99 points better. On an OpenForensics-derived benchmark the same instability becomes a misranking: the checkpoint that validation accuracy selects (98.43%) scores 88.85% on a disjoint 10,905-image test partition, rejecting 21.1% of genuine photographs, while the terminal average, ranked last on validation, scores 94.76%. A study of twenty-two in-house configurations on the identical DF-Wild-Cup split, including preprocessing ablations, shows that CLIP pre-training rather than ensembling drives accuracy, that secondary face cropping costs about 34 points, and that frequency-domain inputs transfer poorly. We recommend per-class selection criteria measured off the early-stopping split.

Deepfake detection models achieve high overall accuracy but exhibit large performance disparities across demographic groups, causing some groups to experience substantially higher false-positive rates than others. Existing fairness-aware methods address this by constraining model outputs to equalize group predictions, but do not tackle the underlying feature-level bias. As a result, models may still rely on demographic characteristics, such as skin texture, as indirect cues for detecting manipulated content. To address this problem, we propose Intersectional Invariant Learning (IIL), a framework that improves feature learning rather than only adjusting model outputs. IIL alternates between identifying the most challenging demographic partitions and learning features that remain consistent across those partitions. The framework also includes a false-positive-rate fairness objective and group-balanced sampling across six gender x skin-tone intersectional groups. We evaluate IIL on a large-scale fairness benchmark containing over one million images, 10 baseline detectors, and 21 fairness and utility metrics. Results show that IIL achieves Pareto-optimal performance, simultaneously attaining an AUC above 99%, a false-positive rate under 2.5%, and the lowest intersectional fairness disparity (FMEO) among all compared methods. IIL outperforms prior fairness-aware methods on the large majority of the 21 evaluation metrics, while reducing both false-positive rates and intersectional fairness disparities by several-fold compared with the strongest competing method.

Deepfake detectors routinely achieve high aggregate accuracy while exhibiting substantial demographic disparities in their errors: false-positive rates differ by double digits across gender and skin tone. Most existing remedies require either full retraining with a fairness-aware objective or demographic-annotated training data, both impractical when only a pretrained checkpoint is available, while label-free post-hoc calibrators leave the detector's weights unchanged. We introduce FYC (FPR-disparity Weight Perturbation and Youden Calibration), a two-stage post-hoc method that perturbs a detector's weights under an FPR-disparity criterion, then calibrates a separate decision threshold for each demographic group - requiring no architectural changes and no full retraining. Evaluated against every detector benchmarked on AI-Face v2 across six intersectional gender x skin-tone groups, FYC outperforms the field on 18 of 21 reported fairness and utility metrics, including purpose-built fairness architectures, at a fraction of the training cost. We further identify why the choice of fairness criterion is critical: the original Demographic Parity objective used in prior weight-perturbation work is architecture-dependent and unpredictable, improving two of four evaluated architectures, leaving a third largely unchanged, and substantially degrading the fourth (PG-FDD), with no way to anticipate the outcome for a given checkpoint in advance. Conditioning the same objective on real images only - our FPR-disparity criterion - eliminates this unpredictability, improving fairness and accuracy jointly across all four architectures with no exceptions.

Vision-Language-Action (VLA) models offer a promising paradigm for general-purpose household robotics. However, inference and data-collection strategies that have been deployed in smaller-scale settings do not necessarily generalize to large-scale, diverse task suites. To address this gap, we conduct a systematic, paired evaluation of previously proposed inference and data-collection strategies across all 100 tasks. Our evaluation uses a fine-tuned $π_{0.5}$ model based on the 2025 BEHAVIOR Challenge winner, which achieves a Q-score of 0.39 on the public test set. These results show that proposed strategies do not transfer uniformly to new environments and tasks. Project page, videos, and detailed analysis: https://rubenhammele.github.io/behavior2026

Real-world forgery is inherently heterogeneous: it manifests not only as digital manipulation but also as physical spoofing, and is carried not only by images but also by videos. Despite substantial progress, a unified anti-forgery foundation model covering such a broad threat spectrum remains absent. In this paper, we present Venus-DefakerOne2, which substantially extends the forensic capability boundary along both the forgery-form and media-type dimensions. Trained on tens of millions of open-source and in-house samples, it incorporates two key designs. Tide-Style-Tuning alternates LoRA and full-parameter fine-tuning to balance computational efficiency, generalization, and hard-sample fitting while mitigating catastrophic forgetting. Entropy-based Test Adaptation determines the optimal prediction modality by analyzing prediction entropy. Venus-DefakerOne2 achieves new state-of-the-art performance across a comprehensive suite of benchmarks, including DeepFake faces (96.7%), AI-generated images (93.3%), tampered documents (87.6%), manipulated natural images (89.6%), physical spoofs (93.1%), and AI-generated videos (96.0%). It significantly outperforms both specialized expert models and its predecessor, DefakerOne, establishing itself as a powerful and versatile foundation model for binary-decision forensic tasks.

A widely held assumption in knowledge distillation is that teacher and student representations must occupy a compatible or explicitly aligned feature space. We propose RPKD (Random Prototype Knowledge Distillation), which discards this assumption by comparing both networks against a shared bank of randomly initialized, frozen prototype vectors. Alignment is therefore carried out in an external reference space rather than in the teacher's native feature coordinates: the teacher side is projected by fixed random maps that carry no trainable parameters, and only the student adapts. Two branches handle the transfer: a logit branch aligning global prototype similarity distributions and a feature branch matching spatially-resolved prototype activation maps. Both operate within the same fixed prototype space, which can be interpreted as a random feature embedding that approximately preserves inter-sample similarity structure, so the prototype bank and the distillation objective stay unchanged across teacher-student pairs while only the input projections are instantiated per architecture. In low-category settings, decoupling the prototype vocabulary from the task label space yields richer supervision than the label space alone, an advantage absent from logit-based methods, whose supervisory signal collapses as class count shrinks. This advantage is especially relevant in real-world scenarios where the label space is typically constrained. Extensive experiments demonstrate the effectiveness of RPKD, achieving a maximum gain of +7.98% over OFA on CIFAR-100, +12.44% on ImageNet-100, and +1.28% on ImageNet-1K. Source code is available at https://github.com/HamzaEtcibasi/RPKD

Implicit neural representation (INR) methods for deformable medical image registration require the network parameters to be optimized for each new image pair, making adaptation speed a practical limitation. This work investigates whether early adaptation can be accelerated using an initialization learned from a collection of previously solved registration tasks. We examine two schemes: Reptile meta-initialization and STRAINER transfer of the shared component of a coordinate network. KAN-IDIR is used as the base INR displacement-field model. Experiments are conducted on OASIS-1, with 364 consecutive pairs used for training and 49 pairs for test-time adaptation. Registration quality is evaluated using mean Dice and HD95, together with the fraction of voxels with a negative Jacobian determinant at iteration 1500. In the configurations studied, both schemes yield higher Dice and lower HD95 at early iterations than random initialization. Reptile numerically outperforms STRAINER under a small iteration budget, although their pretraining budgets are not matched. As the number of iterations increases, the differences between the schemes shrink. STRAINER pretraining takes 39-40 minutes, compared with 44.98 hours for Reptile.

We address the task of long-video, camera-controlled generation from a single image. The dominant autoregressive paradigm suffers from two key limitations: (1) error accumulation over time, as distant frames are conditioned on previously generated outputs rather than ground truth, leading to drift; and (2) ambiguity in how to incorporate camera control. Standard solutions to both--such as self-forcing and conditioning on explicit 6-DOF poses--either incur significant computational overhead or produce drifting scene geometries over time. We propose a simple and lightweight framework that combines autoregressive generation (WanAR) with hierarchical long-context denoising (FixAR). To enforce scene geometry remains consistent with input cameras, we draw from feed-forward novel-view synthesis and reformulate video generation as a warping-and-inpainting problem. To mitigate drift, we introduce a hierarchical denoising strategy in which different stages condition on different temporal histories. Specifically, we first generate frames conditioned on the immediate past, and then refine them using a second denoiser conditioned on a temporally subsampled long-range history to enforce global consistency. This fine-to-coarse scheme enables each generated frame to effectively condition on the full temporal context without incurring a prohibitive cost. Experiments on RealEstate10k and DL3DV demonstrate that WanAR+FixAR outperforms prior methods for camera-controlled long-video generation, producing results that are both temporally stable and consistent with input camera trajectories.

Neural representation has emerged as a powerful tool to mitigate the demand for increasingly higher-resolution images. However, a fundamental trade-off remains between structured (e.g., feature grids, which offer spatial consistency) and unstructured (e.g., 2D Gaussians, which adaptively capture local variances) representations. This choice significantly affects how well an algorithm represents an image. Here, we present Gaussian-Grid, a hybrid neural representation that adaptively integrates structured and unstructured primitives. By dynamically partitioning computation between 2D Gaussian distributions and their attached multi-resolution feature grids based on the local frequency content, Gaussian-Grid automatically optimizes its architecture to match the underlying complexity of the target signal. We evaluate our approach using high-resolution natural images and demonstrate its applications as GPU-native dataset storage, showing $1.24$-$2.14\times$ decoding speed improvement and PCIe data traffic reduction of $23\times$-$176\times$ over JPEG + ZIP data archiving. The results show that Gaussian-Grid maintains small memory footprints while achieving high reconstruction fidelity across a variety of datasets.

Multimodal Large Language Models (MLLMs) ground language in visual evidence, yet their semantic reasoning can be exploited through transferable black-box adversarial attacks. Existing attacks typically allocate perturbations using surrogate-model saliency, which may not reflect how an unseen MLLM semantically interprets or relies on different image regions. We introduce VISTA (Vulnerability-Informed Semantic Transfer Attacks), a black-box framework that identifies semantically vulnerable visual regions from an MLLM's responses without access to gradients, logits, parameters, or internal representations. VISTA supports two attack settings: (i) VISTA-Proxy, a pure-transfer setting that probes an open-source MLLM and transfers the resulting adversarial examples to unseen commercial MLLMs without querying them; and (ii) VISTA-Victim, a query-assisted setting that probes the target MLLM to capture model-specific semantic vulnerabilities. VISTA builds an Uncertainty-Based Multimodal Behavior (UBMB) map from response drift, conflict, misleading, and semantic shifts, then uses entropy-adaptive Sinkhorn routing and a surrogate ensemble to focus perturbations on vulnerable regions. Across GPT-5, Gemini-3.5-Flash, and Claude Sonnet 4.6, VISTA achieves up to 97 % attack success rate (ASR) while maintaining a Peak Signal-to-Noise Ratio (PSNR) above 30 dB. Notably, VISTA-Proxy achieves up to 93 % ASR without querying the final victim, demonstrating that response-derived semantic vulnerability can enable effective transfer across heterogeneous MLLMs.

Ball bounces and racket hits are the events from which tennis statistics are derived, and on monocular video the two are easy to confuse. We release a benchmark of 345 bounce and hit annotations at frame level on eight clips that mix broadcast, fixed elevated, phone and court-level viewpoints, with a frozen development and test split, a confusion-aware matcher, clip-level bootstrap intervals and a blind second annotation of three clips. On it we score supervised precise spotters with published broadcast weights, two zero-shot vision-language models, and RallyArb, an arbitration layer over a ball track and player poses that fits no parameter on event labels. The supervised spotters score near-perfectly on broadcast footage and almost nothing on phone and court-level footage, where the pooled bounce F1 of E2E-Spot is 0.130. RallyArb keeps a bounce F1 above 0.8 on the three phone and court-level clips, two of which were used to tune it, and finds 8 of 11 bounces on the held-out one, but it places hits too far from the contact frame to match human timing. The vision-language models detect events on every viewpoint and miss the contact frame by about 0.1 s. Fine-tuned leave-one-clip-out from its broadcast weights, the spotter reaches about 0.8 bounce F1 on two phone and court-level clips and stays at 0.2 on a grazing phone angle. The annotations, the evaluator and the frozen predictions of every scored row are released.

Objectives: Agencies invest in bicycle facilities but lack tools to assess whether cyclists use them as designed. We examine whether usage patterns follow facility configurations and whether higher bike lane usage is associated with less wrong-way riding. Methods: Motion-triggered cameras covered 19 deployments at 17 locations in Tempe, Arizona, in 2023. Manual review identified 1,963 cycling events; excluding 20 crossings left 1,943 labeled events for analysis. An automated pipeline on the same captures counted 92.9 percent of the full manual total (per-deployment r = 0.99), and every automated wrong-way call was visually verified. Deployments were classified using thresholds on facility usage shares. Findings: Deployments were bike-lane dominant (9), sidewalk dominant (6), or mixed (4). Cyclists in bike lanes rode against the flow in 4.6 percent of events; on sidewalks, riders opposed the near-side traffic flow in 49.6 percent. Across deployments, bike lane share and wrong-way rate were negatively correlated (Spearman rho = -0.87, p < 0.001, n = 16). An independent volunteer count in the same city showed a similar sidewalk-direction pattern (rho = 0.79). Novelty: The study combines multi-site still-camera observations with a fully hand-verified benchmark to characterize facility use and travel direction. Practical Applications: The workflow offers planners a low-cost screen for locations with high sidewalk cycling and travel against the near-side traffic flow, supporting more detailed site investigation.

Traditional computed tomography (CT) can capture paired 3D face and skull geometry, but it remains costly, fixed in location, and exposes patients to radiation. For applications such as remote diagnostics, digital healthcare, forensic identification, missing-person investigation, and archaeological analysis, there is an urgent need for a portable, safe, and accessible solution to capture 3D skull geometry. In this work, we introduce Vision Tomography (VT), a concept and framework capable of reconstructing paired 3D face-skull geometry from various visual modalities, and also enabling joint manipulation of 3D facial and skull geometry. To achieve this, we first construct a dataset of over 200 samples, where each case includes a CT-based skull, a CT-based face, and a high-fidelity textured face scan. Secondly, we propose a novel dense ray matching registration method that ensures topological consistency across face, skull, and their tissue correspondences. Building on this registration, we introduce the 3D Bidirectional Face-Skull Morphable Model (BFSM) as the foundation of Vision Tomography, which enables shape inference between the face and skull through a shared coefficient space, while also modeling tissue thickness variation to support one-to-many facial reconstructions from the same skull, reflecting individual facial changes such as age or body fat. Finally, we demonstrate the effectiveness of Vision Tomography across various forensic and medical applications, including paired 3D face-skull reconstruction from images or sketches and predicting surgical outcomes with joint face-skull editing. Extensive experiments confirm the robustness and accuracy of our method. Vision Tomography is available at https://github.com/wang-zidu/BFSM

Recent advances in monocular depth estimation have demonstrated highly accurate 3D reconstructions within in-distribution terrestrial domains. This study investigates the efficacy of several state-of-the-art approaches to reconstruct out-of-distribution lunar environments using the NASA Ames Polar Optical Lunar Analog Reconstruction (POLAR) stereo dataset and Apollo 17 surface imagery. By leveraging these datasets, we establish a robust analog-to-field validation pipeline that evaluates quantitative performance against ground-truth physical analogs and qualitative performance against historical field data. Our results indicate that select state-of-the-art techniques can successfully generalize to produce high-fidelity relative 3D reconstructions of the unstructured lunar surface.

Benchmarks such as WildfireSpreadTS (WSTS) have enabled data-driven forecasting of wildfire spread from multimodal satellite time series, but they suffer from limitations: size restricts them to large fires that burn for weeks, and they are from similar environments. We introduce AFSeq, a multimodal dataset of 1,194 recent fire events (2023 to 2026) from 8 different countries across three regions, dominated by the short, small fires that existing benchmarks underrepresent. Our dataset is the largest and most diverse to date for time-series-based wildfire spread prediction. We also introduce UTAE-AR, an extension of a classical U-Net with Temporal Auto Encoder, enhanced with recurrent modules before the decoder. This allows the model to not only predict the next day but the whole sequence in an autoregressive way. Moreover, our model can feed its own predicted fire mask back one day at a time, producing the day-by-day forecasts that single-step models cannot. Our results show that UTAE-AR surpasses state-of-the-art on both WSTS, WSTS+ and AFSeq for next-day accuracy while having 21.5% of their parameters, and being able to predict confidently at a longer horizon.

We introduce The Prestige, a video understanding benchmark that probes cognitive visual reasoning through professional magic performances, where deliberate misdirection challenges models to track events and explain unexpected outcomes. The benchmark comprises open-ended questions ranging from temporal and causal understanding to robustness against adversarial questions. Evaluation of leading open-source and proprietary VLMs reveals a substantial gap between even the strongest model and human performance. We also find systematic discrepancies between human and automated grading and identify recurring model failure modes. Ablations further show that models can select correct multiple-choice answers even when they fail to generate correct explanations independently, motivating further investigations into question formats. Together, these findings highlight limitations in both VLM reasoning and its evaluation, underscoring the need for more reliable video understanding models and more rigorous benchmarks.

LOL-v1 and LOLv2-Real are the standard benchmarks for supervised low-light image enhancement, and several leading methods ship hand-set test-time adjustments with their evaluation scripts. We audit both benchmarks with 18 public checkpoints of eight methods under one metric implementation that reproduces every published number to within 0.09 dB, and find that their test labels differ from their training labels. (i) The 15 LOL-v1 test labels are rendered differently: the same captures appear in LOLv2 with labels of identical luminance but 1.48x lower saturation, and LOL-v1's own training labels match the LOLv2 rendering; we confirm this on the official release. (ii) The LOLv2-Real test set is the contiguous block of images 690-789, whose labels are 14% darker than the training labels (Cliff's delta = -0.45, p < 1e-12) with no comparable shift in the input statistics we examined; methods trained without any labels score up to 5.5 dB higher on test than on train. (iii) All 15 LOL-v1 test inputs reappear in LOLv2-Real, 12 in its training set. The released test-time settings are consistent with compensating for these shifts. Separating what can be calibrated on training data from what only test-set tuning finds, the test-only residual has median 0.00 dB on LOLv2-Syn, which has no shift, but 0.37 dB on LOL-v1 and 0.92 dB on LOLv2-Real (up to 3.9 dB). Under the shift, test-set checkpoint selection inflates a from-scratch model by 0.9-2.0 dB, and removing the released settings changes which method ranks first. We give concrete reporting recommendations.

Controllable human motion generation has largely focused on specifying what action should occur or where the body should move. We study a complementary question: can full-body motion be controlled by specifying only when each foot contacts the ground? We introduce StepControl, an autoregressive diffusion model conditioned on a future binary left/right foot-contact schedule, without requiring target root and body poses, or semantic prompts. StepControl represents contact as explicit temporal events and injects them through dedicated cross-attention with classifier-free guidance. We further introduce a stochastic mixture noise schedule and warmup strategy for stable long-horizon autoregressive generation. Experiments show that StepControl follows prescribed contact timing more accurately than adapting existing spatially guided motion generators to contact constraints. It maintains effective contact control over long continuous rollouts and responds to edited schedules such as cadence changes, stop-and-go pauses, limping gaits, and bilateral jumping. Our results show that minimal foot-contact schedules provide an effective and intuitive temporal interface for controllable motion generation.

Video understanding keeps moving toward longer recordings, and the longer a recording is, the more questions can be asked about it. But processing it again for every question does not scale, so many systems watch the video once, build a memory, and answer later questions from it. Yet what to remember is decided before anyone knows what will be asked, and whatever is left out cannot be recovered later. Moreover, whether the savings pay off is judged by token counts, which fall when the input gets shorter but not when the model reuses what it has already computed. We present REEL, an evaluation framework that compares three saved representations built from exactly the same observed frames, so that a change in answers can come only from what the system remembered. Text writes a short record of what happened, Visual keeps the encoder outputs, and Prefix State also keeps the language-model prefix. Each is compared with full reprocessing (Frames) by paired correctness changes, cumulative GPU time, and storage size. Across six vision-language models and two benchmarks, Text and Frames differ in correctness on 20\% to 29\% of questions, even when their mean accuracy is almost the same. By contrast, Visual and Prefix State stay much closer to Frames. With unlimited storage, Prefix State is the cheapest in total GPU time among memories that keep almost all of the answers Frames gets right, for every model on both benchmarks. The trade-off is storage: Prefix State keeps up to about 21 GiB per video, whereas Text needs only kilobytes. Reusing computation and discarding information both look like efficiency, but only the second removes evidence that later questions may need, and neither mean accuracy nor token count tells them apart. REEL makes that difference measurable for any system that answers many questions from one saved memory.

Building damage assessment (BDA) from very-high-resolution imagery supports disaster response, and deep models now achieve high overall accuracy on public benchmarks. Because damage is spatially clustered, most buildings share the damage level of their neighbours, so overall scores are dominated by such buildings, whereas damaged buildings among intact ones are rare and easily overlooked. A high overall score therefore cannot tell whether a model judges the building itself or relies on its neighbourhood as a shortcut, and this study tests which of the two current models do. We stratify buildings by their agreement with their neighbours, move buildings into new neighbourhoods with their pixels unchanged, and diagnose nine models covering the main paradigms of BDA. All models miss a third to half of the isolated damaged buildings, and in non-flood events they call 73--87% of intact buildings damaged once these are moved into a damaged neighbourhood. We propose a building-evidence plug-in that adds two experts to an existing model without retraining it. One expert sees only the building, and the other sees its surroundings but is trained with seamlessly swapped neighbourhoods so that it learns not to trust them. With each of the nine models, the plug-in maintains or improves overall accuracy, reduces errors on buildings that disagree with their neighbours by 8--18%, and cuts by 40--69% the intact buildings that a damaged neighbourhood turns into damaged ones. It also raises the detection of synthetic isolated damage from 1--6% to 13--39% and carries over to optical--SAR data and to an unseen earthquake. High overall scores thus conceal a systematic reliance on the neighbourhood, which building-level evidence reduces at low cost.

Federated Learning (FL) allows for secure analysis of distributed data without data sharing. We and others have successfully applied FL to neuroimaging tasks, such as Alzheimer's disease classification from structural magnetic resonance imaging (MRI). FL trains a deep learning model locally at each site, encrypts the local model parameters and sends them to a federation controller that aggregates the local models under homomorphic encryption, the resulting community model is sent back to the sites for further training, and the process repeats for a number of rounds. FL often achieves learning performance (e.g., accuracy) close to centralized training, but no data leaves the sites and model parameters are protected through encryption and other measures. In this paper, we show that federations composed of sites with heterogeneous amounts of data (1 large site plus small sites) perform better than homogeneous (same data per site) federations, particularly in low data regimes, while reaching performance close to centralized training. Having a big site often results in faster convergence and higher final accuracy. Heterogeneous federations better exploit sites with small amounts of data, a common situation in large consortia such as ENIGMA.

Short video platforms have become a primary channel for black and gray industry operations in internet finance, where inducements such as prompts to like, follow, join private chat groups, or click external links frequently steer users toward fraud. Because such videos are short, dynamic, and often convey intent through fleeting visual cues rather than stable text, a moderation system must guarantee detection recall and localize the offending segment so reviewers inspect only the relevant clip. Existing OCR+LLM pipelines transcribe on-screen text frame by frame and classify the transcript, but exploit only textual information. Many inducements are conveyed through visual cues OCR cannot observe, and keyframe-based localization cannot recover precise temporal boundaries. We explore OCR+MLLM, which augments the text pipeline with a Multimodal Large Language Model, and compare three routes: OCR+LLM, OCR+MLLM, and a large zero-shot MLLM, along detection, latency, token consumption, and deployment complexity. OCR preserves the text signal that guarantees recall, while the MLLM jointly performs OCR-grounded discrimination and temporal grounding, localizing both textual and non-textual inducements. We construct a dataset of 2,419 internet finance short videos (66.49 hours) with 3,194 annotated events across four categories, and study reward design for RL-based grounding under class imbalance. We report two findings: auxiliary reward terms are exploited through reward hacking, and stronger localization rewards degrade class-balanced detection. We therefore design a recall-weighted F2 localization reward with class-balanced resampling. Experiments show OCR+MLLM localizes better than the keyframe OCR+LLM baseline (mIoU 26.52 vs. 14.49) and detects non-textual inducements that text-only pipelines miss, while maintaining 67.0 percent macro recall.

Walking offers a window into physical function that may support interpretable video-based frailty assessment. We developed silhouette-based "gaitomics" to connect frailty predictions with measurable patterns of body configuration and movement. In 66 older adults assessed using the Fried Frailty Phenotype, standardized silhouettes and gait energy images yielded 90 descriptors of shape, temporal variation, motion, and texture. Minimum redundancy maximum relevance selection using the mutual information difference criterion prioritized informative features while limiting redundancy. Each classifier was evaluated using nested stratified cross-validation with five outer and three inner folds at the participant level, with feature ranking and feature-count selection confined to training partitions. Shapley additive explanations examined contributions to held-out predictions. Bilateral silhouette difference led the aggregate feature-selection rankings for three-class classification and both comparisons involving frail participants. Attribution profiles differed across tasks, with texture and consecutive-frame silhouette differences prominent in three-class classification. Among classifier-specific evaluations, CatBoost achieved the highest mean three-class micro-AUC of 0.647 \pm 0.085. Leading pairwise mean micro-AUCs were 0.833 \pm 0.249 for frail versus nonfrail, 0.765 \pm 0.168 for frail versus prefrail, and 0.554 \pm 0.129 for prefrail versus nonfrail classification. These findings connect model predictions to explicit gait characteristics while revealing task-dependent feature priorities and limited separation of prefrail and nonfrail participants. The framework provides an interpretable basis for further investigation, with larger independent studies needed to establish reliability and clinical relevance.

Language models increasingly create 3D content through code, with practical workflows often relying on editing existing programs rather than regenerating them, for speed and precise control. Such edits require understanding what a program builds and foreseeing how a change reshapes it, yet prior benchmarks mostly evaluate generation or simple edits against a reference. In this paper, we introduce 3DCodeEdit, a benchmark of 2,592 instruction-guided editing tasks over 108 Blender programs that construct objects entirely in code. Its edits range from parameter to structural changes, style redesign, and scene composition, and are judged from renders against human-reviewed checklists. Across 17 models, the average success rate is only 37%, and even the strongest models complete fewer than 60% of structural and style edits. As models become more capable, their failures shift to geometric foresight, where the code runs but builds the wrong shape. We therefore study whether multi-turn feedback can compensate for this missing foresight. Feedback from a separate reviewer model that describes how the renders differ from the request helps most, more than maximum reasoning effort or an agent harness, and can let a 27B model surpass the single-shot rate of a 2.8T-parameter model.

Current datasets for text-to-video generation are largely limited to short clips depicting a single event or coarse-grained captions. As a result, they provide limited supervision for modeling long-take temporal dependencies, often leading to temporally inconsistent generations when describing multiple events. In this paper, we introduce MEV, a multi-event video dataset designed for long-take video generation. MEV provides high-quality real-world videos where subjects can perform a sequence of distinct, non-overlapping events, featuring precise temporal boundaries and fine-grained procedural annotations. This event-centric discretization reduces confounding factors and enables models to support temporally localized conditioning and build long-take videos from atomic event units. Furthermore, we introduce a comprehensive annotation pipeline that segments videos into temporally grounded events, and generates structured captions at both the global level and the event level. In addition, under an event-centric evaluation protocol, we conduct multi-event generation experiments by fine-tuning on MEV with temporally grounded event conditioning, leading to improved long-take event transitions and more coherent content. We believe this work provides a publicly available resource for multi-event fine-tuning and event-centric evaluation, together with practical insights into learning from temporally structured video-text data.

Topology-agnostic motion generation aims to synthesize motions for diverse characters with arbitrary skeletons. Unlike conventional motion generation, which assumes a fixed skeletal template, this setting must handle diverse joint configurations and inconsistent joint semantics across characters. Existing methods address this challenge through dense full-joint modeling, which incurs substantial computational cost and hinders learning transferable motion patterns across diverse skeletal topologies. We present ETA-Motion, a topology-agnostic motion generation framework based on sparse joint learning. Instead of learning motion from all joints, our method uses a compact subset of joints to reduce redundancy and trains the motion generator on the resulting sparse skeleton. To facilitate correspondence across diverse topologies, we introduce semantic joint embeddings that combine textual and visual cues, together with topology-aware training objectives for preserving skeletal structure. Although ETA-Motion is trained only on sparse skeletal observations, it supports full-skeleton motion generation during inference. Experimental results demonstrate that ETA-Motion outperforms existing topology-agnostic baselines while significantly improving efficiency and generalization to unseen motions and unseen topologies.

Data augmentation improves visual recognition by exposing models to synthetic training examples that encourage generalization beyond the observed data. Mixed-sample methods extend this idea by combining multiple images. However, many common approaches either mix only pairs of images or rely on fixed small layouts, limiting control over source count, visible area proportions, and induced supervision. We propose TreemapMix, a distribution-controlled multi-image augmentation that samples source weights from a Dirichlet distribution and assigns them to area-proportional regions using a treemap partition. The resulting known region areas support two complementary objectives. TreemapMix with soft cross-entropy (SCE) treats areas as soft class-probability targets, while TreemapMix with a Plackett-Luce (PL) objective converts areas into ordinal supervision. On ImageNet-1K under matched training recipes, TreemapMix-SCE achieves substantially lower calibration error than evaluated mixed-sample baselines, while TreemapMix-PL achieves the strongest Top-1 accuracy. Transfer experiments on object detection and instance segmentation suggest that TreemapMix-pretrained features remain useful beyond classification. These results show that controlled multi-image composition can expose the same region-area information as either probability-matching or rank-matching supervision. Code is available at https://github.com/Ejafa/treemapmix.

Feed-forward 3D Gaussian Splatting has recently emerged as a promising alternative to per-scene optimization, enabling fast novel view synthesis by predicting Gaussian primitives directly from input images. However, it quickly faces a bottleneck of rendering efficiency while scaling from sparse to dense multi-view inputs, which offer more geometric constraints but produce a rapidly growing set of candidate Gaussians. Existing solutions cannot reliably distinguish redundant primitives, leading to many redundant primitives from repeatedly observed regions and increased rendering cost without much gain in visual quality. We propose BudSplat, a compact feed-forward 3DGS framework that learns to keep the most valuable Gaussians under a limited primitive count. BudSplat has two novel designs. The first is a detail-aware importance score that is extracted from multi-level visual tokens, providing an explicit cue for regions where pruning is likely to harm perceptual quality. The score can be injected into voxel fusion to preserve high-value local contributors during multi-view aggregation. The second is quality-aware Gaussian pruning that selects a compact set of primitives by jointly considering rendering contribution and detail importance. The quality-aware pruning enables local quota redistribution that helps avoid over-allocating primitives to large redundant regions. Experiments on RE10K, DL3DV, and ACID show that BudSplat achieves a favorable trade-off among rendering quality and efficiency.

Vision Transformers (ViTs) achieve strong image recognition performance, but their self-attention increases quadratically with the number of input tokens and makes them computationally expensive. The standard implementation of ViT models treats all patch tokens with equal importance across every layer of the model. This uniform treatment introduces redundant computation and may reduce the model's focus on the most discriminative object regions. In this paper, we propose ConvRoute-ViT, a convolutional token routing mechanism for efficient vision transformers that performs token reduction through convolutional routing. Our model preserves the most relevant patch tokens according to class-token attention at selected layers, while the remaining tokens are spatially reorganized and compressed into a compact set of convolutional summary tokens. This design reduces the sequence length without completely removing the contextual information of the less informative tokens. By applying the routing process hierarchically, the proposed framework constructs compact and discriminative representations for deeper layers. Experiments on CIFAR-100 and TinyImageNet show that ConvRoute-ViT improves computational efficiency while maintaining classification accuracy compared with the base ViT.

Pneumonia changes across respiration, inflammation, and organ function, while the measurements that describe it arrive at different times and are often incomplete. We introduce PULMOWORLD, a world model for longitudinal pneumonia dynamics that connects asynchronous state estimation, recursive simulation, and evidence-constrained planning. History-aware filtering separates state propagation from observation-driven correction, so a measurement updates the shared belief only when it is observed. That belief supports recursive forecasts of clinical values and modality representations, and summaries of future prognosis. Cap-MBRL searches imagined continuations inside an authorized action set and replaces a frozen fallback only when a candidate meets the risk budget and a predicted-cost margin. On the pneumonia cohort MASS, recursive forecasts are competitive with task-trained sequence models, and simulated future states retain prognostic information. Trained separately on public MIMIC-IV-CXR chest radiographs and clinical trajectories, this model improves prognostic discrimination over latest-value fusion in development and is competitive on long-horizon error on the locked pneumonia panel. In five known-mechanism MASS-Syn environments, Cap-MBRL lowers selection regret relative to direct fallback execution.

Language-model answers are often treated as a visible record of the internal decision that produced them. We test this assumption in conflicting-context question answering, where two supplied documents support incompatible answers. A model can process the disagreement and still return a one-sided answer, even under prompts that ask it to disclose conflict. In Qwen 0.8B, hidden source commitment is captured by a residual source-choice direction whose sign tracks which document the answer uses. Sparse middle-layer MLP writers feed the same coordinate, and steering the coordinate changes the chosen source. A transfer run on Gemma 3 1B preserves the source-choice readout and sparse-write bridge under LLM-audited labels. Conflict disclosure follows a different pattern, appearing weaker, later-layer, and only partially steerable. These results support a practical monitoring strategy for retrieval systems that detects silent source commitment with residual readouts and pairs conflict-surfacing prompts with disclosure steering when the model has selected a side under disagreement.

Large vision-language models (LVLMs) represent images and videos as long visual-token sequences, substantially increasing inference costs. These costs motivate visual-token pruning, whose effectiveness depends on both what signal is used for selection and when that signal becomes available. Our analysis reveals complementary limitations: visual-side [CLS] attention produces scattered, background-biased selections under aggressive compression, while stronger query-conditioned signals become available after substantial computation within the large language model (LLM). This raises a natural question: can structurally coherent, query-aware token selection be performed before consulting the LLM? We therefore propose ST-Pruner (Spectral-Textual Pruner), a training-free framework for pre-LLM visual-token selection. It combines attention-independent spectral saliency with inverted patch-text similarity through parameter-free multiplicative fusion. Across diverse LVLMs, ST-Pruner retains over 95% of full-token performance using at most 11.1% of visual tokens and achieves a 3.2x end-to-end speedup on LLaVA-NeXT. Code is available at https://github.com/sejong-rcv/ST-Pruner.

Football action valuation estimates players' contributions to future scoring. Existing multi-agent deep reinforcement learning (MADRL) models use 2D positions and velocities, omitting posture information relevant to imminent actions. We investigate whether adding 3D pose to a SARSA-based model improves observed-action values relative to attack returns: goals, conceded goals or final expected possession value. Using UEFA Champions League 2022/23 event, tracking and 3D pose data, we compare six representations (no pose, raw joints, torso and head orientation, and spatial-temporal graph convolutional network (ST-GCN) embeddings) across five learning settings and three seeds. The settings vary temporal-difference targets and action-prediction loss, with Monte Carlo return regression as a reference. Across 50 on-ball and off-ball comparisons, pose robustly improved return prediction in only three, all using torso and head orientation, while worsening it in twelve, mostly with ST-GCN. Training choices had greater effects: removing the action-prediction loss raised on-ball skill of position-only values from 0.003 to 0.040 and their correlation with expected goals from 0.23 to 0.60; the improvement in return prediction persisted at matched training length. Separately, pose improved prediction of the on-ball action 0.2 s ahead (macro AUC 0.914 to 0.934 with raw joints), while the passer's pose during execution improved pass-success prediction (AUC 0.844 to 0.856). Within this training protocol, pose informed imminent actions but rarely improved observed-action values.

Extremely sparse-view computed tomography (xSV-CT) is critical when scanning time, motion, or radiation tolerance limits each scan to only a few projections. In many such scenarios, the same acquisition limits also constrain dataset construction, making pre-reconstructed 3D volumes from dense scans unavailable. This exposes a gap in current sparse-view approaches: Optimization-based methods, which use the sparse measurements directly but lack data-driven priors to suppress artifacts, and learning-based methods, which use strong priors but usually require densely scanned training volumes acquired outside the xSV-CT regime. We study this train-sparse / infer-sparse setting by learning a pose-aware diffusion prior from one 2D projection per training sample, then guiding diffusion sampling with measured-view fidelity, shared-volume reprojection, and projection-physics consistency for tomographic reconstruction. On two public CT datasets across 1-, 4-, and 8-view settings, the proposed method achieves the best novel-view synthesis scores and best or competitive reconstruction quality compared with baselines that use dense-volume supervision.

Perception tasks for navigation in robotics, including aerial platforms such as drones and autonomous driving systems, are inherently structured. Drone-mounted cameras typically capture sky above, terrain below, and obstacles or man-made structures in between, while driving data often contains organized road layouts, lane markings, and surrounding agents. Motivated by these axis-aligned structural priors, we note that such information is typically more structured than in generic image tasks. We hypothesize that processing information in a quadtree-esque manner can not only model features effectively in a hierarchical manner, but also offers an efficient linear-time alternative to vanilla attention mechanisms, which run in quadratic time. In this paper, we propose Sibling-Selective Quadtree Attention (SSQA), which models image tokens hierarchically as a structured, full quadtree. We show analytical complexity analysis that guarantees linear-time feature modeling, in addition to empirical experiments comparing inference speeds with other popular modeling approaches, such as Mamba 2 and Quadtree Attention. Our results, benchmarked across several tasks, show that we achieve results at least as good, if not notably better, as others at a fraction of the computational costs.

Visual foundation models learn rich pathological morphology but lack explicit concept-level supervision. Vision--language pretraining can provide this supervision but requires costly image--text pairs. Here, we introduce GloSAGE, a semantic self-distillation framework that uses lesion profiles predicted by a pretrained visual teacher to construct structured pathology descriptions. Reliability-guided selection identifies image--text anchors, while Knowledge-Structured Pathology Text Synthesis and Evidence-Guided Negative-pair Construction account for partially observed lesions. Image--text alignment trains projection layers while retaining the frozen visual encoder; an intermediate Semantic Transfer Layer supplies features for downstream tasks. Developed on GloMillion and evaluated across seven glomerulus- and case-level cohorts, GloSAGE supported zero-shot lesion recognition, image--text retrieval, hard-negative discrimination and concept-intervention analyses. It outperformed its source visual model in most lesion-recognition tasks and improved lesion grading, while supporting competitive case-level disease classification, injury grading and baseline renal function estimation. Ablations supported the combined contributions of knowledge-structured text and evidence-guided negatives, and the intermediate transfer layer improved mean lesion-recognition ROC-AUC relative to the complete projection output. A frozen language model conditioned on GloSAGE features generated glomerular descriptions. These findings suggest that teacher-derived lesion profiles can provide annotation-efficient semantic supervision for extending a glomerular visual foundation model.

While single-view novel view synthesis methods have been dramatically enhanced recently, how to quantify their performance has yet to be standardized. In this work, we demonstrate that an important factor lies in how each method's own reference frame is aligned with the so-called ground-truth pose. Existing methods often rely on test-time optimization of target poses for this alignment, which inherently is ill-posed due to the complexity of the optimization landscape of camera poses vs render accuracy, leading to methods being evaluated at an arbitrary local optimum. Some recognize this limitation and thus rely on 3D reconstructions of entire scenes using training and test views to align poses, which may be infeasible. In this work, instead, we propose to use traditional 2D-to-2D and 2D-to-3D image matching techniques to align 3D reconstructions. We also evaluate only within co-visible regions, between different views, which is the only place where a `correct' ground truth can be defined. Our evaluations show that with proper coordinate alignment, novel-view synthesis methods that use explicit 3D Gaussians achieve the state of the art, despite being previously reported in the literature to be outperformed by pure image-based 2D methods. We further provide a simple baseline that directly utilizes an off-the-shelf video generator and a feed-forward geometry estimator. Our simple baseline is already comparable to state-of-the-art methods specifically trained for novel-view synthesis, without any task-specific training.

Producing long-form videos from narration requires interdependent decisions that remain coherent across shots. Existing agentic editors often use multi-stage workflows driven by general models, leaving the quality-cost trade-off of compact, task-adapted policies unclear. We address this gap with WeVisCut, a long-form script-to-video system that connects tool execution with policy learning. Its agent harness maintains shared production state so that tool feedback informs subsequent decisions and repairs. Because finished videos do not expose these decisions, our four-stage data pipeline combines editing inversion with verified replay to turn recovered editing intent into executable supervision. Applied to 3,750 finished videos, this pipeline yields over 130K training examples. We use this supervision to train a single 9B policy that coordinates production and recovers from failures within the same harness. Moreover, we introduce a 100-task benchmark and five metrics assessing editing quality and delivery reliability. Under a shared tool environment, WeVisCut substantially outperforms its base model and achieves comparable performance to Claude Opus 5 while using nearly 33x fewer generated tokens per successful output minute. It also outperforms task-matched native editors on our proposed metrics.

3D reconstruction from single-view sports broadcast is a pivotal capability for automated analytics and immersive AR/VR replays, yet notoriously ill-posed due to difficult-to-calibrate moving cameras and depth ambiguity. Unlike widely studied racket sports, team sports such as volleyball feature unpredictable multi-contact sequences and high altitude jumps. These dynamics categorically invalidate the standard planar projection assumptions relied upon by existing methods. To resolve these limitations without 3D training supervision, we propose a complete, physics-grounded pipeline tailored to monocular broadcast volleyball. First, we include a correction module to the perspective drift induced by ground-plane projection by coupling temporal jump spotting with Newtonian interpolation for 3D player placement. Second, for 3D ball trajectory estimation, we design a dedicated sub-trajectory segmentation architecture alongside a collision-matching algorithm. By anchoring physical equations to reconstructed 3D hand vertices, we lift noisy ball 2D detections into a spatiotemporally coherent 3D scene. Evaluations on multi-camera benchmarks demonstrate our approach significantly reduces 3D spatial drift compared to 2D-reprojection baselines.

Sparse-view cone-beam computed tomography is widely used for fast, low-dose volumetric imaging and lesion localization in dentistry and image-guided interventions, but it introduces severe artifacts. Feed-forward Gaussian methods transfer the reconstruction cost to training and decode each coordinate from projection-aligned features and neighboring anchors. However, information from the whole scan is written only when the field is constructed and cannot be retrieved during coordinate decoding, leaving the volume inconsistent within the scan. This is a common limitation of such methods: once the field is fixed, decoding uses only local features and reconstruction quality drops substantially. We propose SparseTokenGS, a Gaussian-based method for sparse-view cone-beam computed tomography reconstruction via a reusable token pool. Anchor Token Fusion and Point-Token Attention query this shared pool at the anchors and at the query coordinates, respectively. The local Gaussian prior is retained, and the volume is reconstructed in one forward pass, without per-scan optimization or a projection-domain loss. On three public chest and dental datasets, SparseTokenGS achieves the highest reconstruction performance among the compared methods. On clinical dental scans, reconstruction from only six projections reaches 29.51 decibels in peak signal-to-noise ratio and a structural similarity of 0.842 against the full-view reference, and off-the-shelf tooth-and-pulp segmentation keeps these structures separable for root-canal localization. Experiments on volcanic rock and chicken show performance after same-object adaptation. A reusable token pool gives feed-forward reconstruction a mechanism that keeps scan-wide context available through decoding and can benefit future rapid clinical lesion localization and analysis. The code is available at: https://github.com/FrankDengAI/SparseTokenGS.

Medical vision-language models (VLMs) show strong performance in radiology report generation (RRG) and medical visual question answering (Med-VQA), producing free-form clinical descriptions and answers from medical images. However, they may generate clinically meaningful statements that lack visual evidence or contradict image content. We refer to these errors as visually irrelevant hallucinations and investigate them at the phrase level. We propose CLEAR (Counterfactual Localized Evaluation of Alignment and Response), a model-agnostic post-hoc framework for phrase-level hallucination risk estimation. CLEAR decomposes generated responses into visually assessable clinical phrases. For each phrase, phrase-conditioned grounding identifies a candidate visual region, which is selectively perturbed by a localized counterfactual intervention. CLEAR characterizes the intervention using two complementary auditing cues: response cues, capturing semantic and clinically meaningful changes in the generated response, and alignment cues, measuring changes in image-phrase compatibility. These cues are normalized and jointly modeled to estimate phrase-level hallucination risk. Across four datasets spanning RRG and Med-VQA, CLEAR outperforms state-of-the-art hallucination detection baselines across multiple medical and general-purpose VLMs. On MIMIC-CXR, CLEAR improves AUPRC from 85.10% to 90.23%. Its effectiveness is further validated on a GT-derived controlled VIH benchmark. These results establish visually irrelevant hallucination as an important fine-grained reliability failure mode in medical VLMs and demonstrate the value of model-agnostic visual-support auditing for more trustworthy multimodal clinical generation.

Real-world deployment of computer vision models, particularly in outdoor scenarios, is often limited by environmental interferences, such as adverse weather. Improving model robustness across weather conditions is thus critical for reliability. However, sourcing high-quality data relevant to isolated weather effects is a significant challenge. Most datasets include changes in dynamic elements across observations of the same scene: subjects such as pedestrians and vehicles change between shots, while camera viewpoints vary. These aspects make it difficult to decouple solely weather-induced distribution shifts from the scene. In this work, we address this with a novel weather synthesis framework that utilizes structure-retaining Low-Rank Adapters (LoRA). We analyze the attention blocks in Stable Diffusion XL (SDXL) and identify where relevant structural information is mostly processed during image-to-image translation. Through an architectural analysis of SDXL, we learn structural LoRA matrices for the corresponding blocks and then leverage them for weather synthesis, eliminating the need for external conditioning such as ControlNet. Experimental results demonstrate that our method improves upon existing synthesis approaches, achieving greater realism while preserving structure. This provides a robust pipeline for generating high-fidelity training and evaluation data for outdoor computer vision applications.

Humans routinely consume multimodal content, where textual and visual information is naturally interleaved. Although Unified Multimodal Models (UMMs) can jointly generate text and images, multimodal interleaved generation remains largely underexplored. Progress in this area is hindered by two key limitations: (1) the lack of well-defined interleaved generation tasks and corresponding training data, and (2) the absence of fine-grained evaluation protocols that can reliably diagnose model performance across different capability dimensions. To this end, we present IntReal, comprising two complementary components: IntReal-100K, a real-world dataset of multimodal interleaved samples spanning four major categories, and the IntReal Evaluation Suite, which evaluates model performance across seven complementary dimensions using specialized expert models, enabling coarse-to-fine diagnosis. Extensive experiments validate our evaluation suite through low inter-metric redundancy (max pairwise absolute Spearman's $ρ\leq 0.42$ ), high agreement with human judgments (83.3% accuracy), and strong consistency across different expert evaluators (Spearman's $ρ> 0.9$ ), supporting the complementarity and reliability of our metrics. By benchmarking four families of UMMs, we show that IntReal Evaluation Suite reveals distinct strengths and weaknesses across models. Post-training experiments demonstrate the utility of IntReal-100K for improving interleaved generation and further show that training unimodal generation can also benefit interleaved generation. These findings suggest that interleaved generation can benefit from modality-specific optimization, rather than requiring joint text-image training at every stage.