New Computer Vision (cs.CV) papers, Friday 23 October 2026
114 papers announced on arXiv. Get them ranked for you, free →
V-JEPA 2 Meets AQA: Phase-Based Scoring and Reference-Based Feedback for Mountain Biking
AI summaryPipeline uses frozen V‑JEPA‑2 video features and pose data to predict binary jump quality; achieves AUROC ~0.77 (0.81 without biomechanical inputs) but shows limited generalization and unreliable feedback, highlighting need for further evaluation.
EgoHand: Monocular 3D Hand Reconstruction from Egocentric Video
AI summaryEgoHand reconstructs 3D hand poses from single egocentric video, trained on 14 M diverse images and using SOMA-Hand with a transformer encoder and camera cues. It outperforms baselines on HOT3D, ARCTIC, and H2O with 9.7 ms latency and gains user preference over competitor models.
TerraTransfer++: Demonstration-Free End-to-End Planner
End-to-end autonomous driving has achieved state-of-the-art performance on benchmarks and real-world deployments. Its standard training recipe, however, is expensive across all existing paradigms: imitation learning requires fleet-scale driving logs whose behavior distribution…
Self-Distilling Future Video Representations for World Action Models
World Action Models (WAMs) have emerged as a promising approach to learning generalist robot policies using rich spatiotemporal priors from pre-trained video diffusion models. A recent line of WAMs improves inference efficiency by conditioning a separate action decoder on…
R4DGS: Referring Segmentation in 4D Gaussian Splatting
AI summaryR4DGS introduces R4D-Bench-QA for referring segmentation in dynamic 4D Gaussian scenes and a training‑free framework, ReferGaussian, that converts Gaussians into language‑addressable entities and uses MLLMs for constraint‑aware grounding, achieving competitive temporal selection and grounding accuracy.
MoRE: Motion-Regularized Encoding for Articulated 3D Generation
AI summaryMoRE encodes articulated 3D objects using joint lines (axis classification and Plücker moments) and motion‑regularized learning, ensuring reconstructed parts follow correct motion. It outperforms baselines on PartNet‑Mobility and ACD in part alignment, state geometry, and joint accuracy.
Foundation4D: Towards a unified perception and reconstruction model for autonomous vehicles
AI summaryFoundation4D unifies geometry estimation, 3D detection, semantic occupancy, and Gaussian reconstruction in a single feed‑forward model for multi‑camera driving video, achieving state‑of‑the‑art depth, detection, and occupancy metrics while improving geometry accuracy through co‑training.
FiLM-CAM: Keyed Feature Modulation for Conditional-Access Watermarking
AI summaryFiLM-CAM introduces keyed feature-wise linear modulation to embed watermarks whose extraction and removal depend on a secret key. The approach achieves a large performance gap between authorized and unauthorized decoding, enabling reliable payload recovery and key‑conditional watermark removal across multiple backbones.
Latent Space Jam: Layout Guided Video Synthesis
AI summaryLatent Space Jam uses a pretrained video diffusion model twice: first to generate a coarse layout guide via cross‑attention, then to produce a high‑quality, layout‑consistent video. It yields stronger temporal consistency, better overlap and trajectory accuracy without extra training.
VisCurate: Evaluating Behavioral Equivalence of Visual Skills
AI summaryVisCurate curates visual skill libraries by testing skill behavior on probe images, assigning six safe relations. It recovers 51.2% of mergeable pairs in benchmarks and shows 92% redundancy in real libraries, outperforming text‑based baselines.
SplatLab: Autonomous Scientific Discovery for 3DGS Research
AI summarySplatLab automatically reproduces 3D Gaussian Splatting baselines, proposes and evaluates extensions, and writes fully traceable manuscripts. It improves PSNR by up to 3.78 dB over published results while compressing research cycles from months to days.
AVAtt: A Free-Viewing Eye-Tracking Dataset and Benchmark for Visual Attention on Paintings
AI summaryAVAtt is a large free‑viewing eye‑tracking dataset of 10k+ paintings, providing scanpaths, fixations, saliency maps, and metadata. Benchmarking shows classic and deep saliency models fall short, highlighting the need for art‑aware attention models.
Unified Trajectory and 3D Pose Forecasting for Multi-Agent Sports Analysis
AI summaryJointly predicts agents’ 3D pose and trajectory using a socio‑temporal encoder with shared attention, per‑joint reweighting, and frequency‑domain smoothing. The model outperforms prior joint forecasting methods on WorldPose and 3DPW, especially improving pose accuracy.
Visual Grounding First, Multimodal In-context Learning Follows
AI summaryProCoRe uses contrastive reinforcement learning on multimodal data to improve visual grounding in multimodal LLMs, boosting ICL classification accuracy by 5.6% and caption ROUGE‑L by 13.8% over Qwen‑VL‑3‑8B without ICL training data.
KinoUAV: A Benchmark for Safe and Stable UAV Vision-Language Navigation with Kinodynamic Modeling
AI summaryKinoUAV presents a kinodynamic benchmark for UAV vision‑language navigation, adding trajectory constraints and directional free‑space distance labels. The KinoUAV‑Agent achieves higher success rates with lower yaw and collision rates by decoupling perception from safe action generation.
GSHeadRelight: Fast Relightability for 3D Gaussian Head Synthesis
AI summaryGSHeadRelight uses generative 3D Gaussian head models with a linear neural radiance transfer for relighting, trained only on in‑the‑wild images. It achieves state‑of‑the‑art relighting quality while rendering 12× faster than prior 3D portrait methods.
Refining Open-Vocabulary Semantic Segmentation via Regional Semantics and Visual Prototypes
AI summaryThis work adapts CLIP’s visual encoder for pixel‑level open‑vocabulary segmentation by adding selective attention to preserve regional semantics and an auxiliary visual‑prototype classifier that refines text‑only predictions. The method improves accuracy across five segmentation datasets.
FedBST: Bilateral Self-Training for Unsupervised Federated Classification
AI summaryFedBST adds bilateral self‑training to unsupervised federated vision‑language learning, using uncertainty‑aware confidence penalization to exploit ambiguous samples and a visual‑text ensemble to calibrate local prototypes. It outperforms prior federated methods on non‑IID benchmarks.
CODI: Improving Self-Supervised Vision Transformers with Cross Distillation
AI summarySelf‑supervised Vision Transformers learn good image‑level features but poor patch‑level ones, because the CLS token focuses on few patches. Cross Distillation (CODI) injects CLS semantics into each patch token, improving dense and image‑level accuracy.
Asking Less, Adapting Better: Feedback-Efficient Test-Time Adaptation for Vision-and-Language Navigation
AI summaryFETA uses a two‑stage selection to ask humans for only the most informative action in a trajectory, then updates the policy at that step with a discrepancy‑aware entropy loss. It outperforms prior test‑time adaptation methods while reducing human queries.
One-shot federated unsupervised domain adaptation with Smoothed Knowledge Distillation and teacher refinement
AI summaryOne-shot federated unsupervised domain adaptation (FUDA) using Smoothed Knowledge Distillation (SKD) and Exponential Moving Average teacher refinement (ETR) trains a student on target data with smoothed pseudo‑labels, then refines source models via EMA, achieving competitive accuracy with a single communication round.
Residual Calibration via Local Feature-Space Refinement
AI summaryThe paper introduces LAPE, a post‑hoc residual calibration technique that refines global calibrator outputs by non‑parametric local evidence in feature space, and extends it to multi‑view inputs via CFAF; experiments show improved calibration for several baselines with low overhead.
Class-Grounded Batik Captioning under Scarce Text Annotation: Label Retention and Source-Specific Error Propagation
AI summaryThis work adapts a LoRA‑tuned DINOv2 classifier to guide Qwen2.5‑VL‑7B batik captioning with scarce text data, achieving a 90.67% class‑label mention rate and 96.10% classifier accuracy while showing error propagation into captions.
AgentDreamer: Agentic 3D Motion Reasoning for Text-Controllable Physics Simulation
AI summaryAgentDreamer uses an agent to translate text instructions into dense 3D motion priors, then optimizes a differentiable MPM simulator to match these priors while respecting scene dynamics. It outperforms video guidance in aligning simulated motion to language, as measured by a new Motion‑IG metric.
Zero-Shot Domain Generalization via Domain Prompt Learning
AI summaryDoP uses a vision‑language model to learn domain prompts from test data only, first generating diverse style tokens in text space (TDL) and then refining them with visual self‑supervision (VDL). It achieves state‑of‑the‑art zero‑shot DG on PACS, VLCS, OfficeHome, DomainNet, and ImageNet.
LoRA Superposition across Pretrained Models
AI summaryLoRA superposition sums LoRAs trained on different ViT backbones for the same task, improving all 56 backbone‑task pairs with a 30.21‑point mean accuracy gain, nearly matching individually trained LoRAs and supporting non‑interfering components.
ParaLin++: Accelerating Parallel Diffusion Integrator via Intrinsic Partially Linear Structure
AI summaryParaLin++ accelerates diffusion sampling by removing solver temporal dependency while keeping full‑step discretization. Experiments show it retains the Euler quality profile, achieving 2.6–4.8× speedups on Stable Diffusion‑v2, LSUN Church, FLUX.1‑dev, and 1.21× on Ming‑Flash‑Omni 2.0.
PhyEvolve: Multi-Agent State-Evolution Planning for Physics-Aware Video Generation
AI summaryPhyEvolve uses a multi-agent framework to decompose user prompts into causally linked physical state transitions for video generation. Without model retraining, this explicit planning approach improves physical consistency and outperforms existing baselines across major benchmarks.
Learning from Next-Frame Prediction: Towards Unified and Scalable Visual Representation Learning
AI summaryVANE trains a causal encoder to predict the embedding of the next video frame, then decodes patches for detail. It scales with data and model size, achieving state‑of‑the‑art results on anticipation, recognition, and dense perception, notably 7.1% mCR@5 better than V‑JEPA on EPIC‑KITCHENS.
SmolVLA-MoE: Sparse Mixture-of-Experts for Efficient Multi-Task Robotic Manipulation
AI summarySmolVLA-MoE introduces a sparse Mixture-of-Experts architecture into lightweight Vision-Language-Action models for robotic manipulation. It achieves an 84.25% average success rate on the LIBERO benchmark, outperforming dense baselines while maintaining high inference efficiency.
Where Did the Light Come From? Diagnosing Cast-Shadow Reasoning in VLMs and Borrowing the Missing Shadow Prior from Image Generation Models
AI summaryThe paper introduces ShadowBench, a real‑photo benchmark testing VLMs on shadow direction, inference, and consistency. It shows that VLMs read visible shadows better with scale but poorly infer absent shadows or judge physical validity; routing VLMs through image‑generation models significantly improves performance, suggesting generation objectives as a path to illumination awareness.
FilmFlow: Agentic System for 3D Virtual Film Cinematography
AI summaryFilmFlow translates natural language or reference videos into executable camera trajectories for dynamic 3D scenes using a Director agent and specialized sub‑agents. On the CinemaBench benchmark, it outperforms baselines and wins user preference, showing superior spatial reasoning and visual composition.
Evaluation of Front-End Feature Matching for Lunar Visual SLAM
AI summaryThe paper benchmarks 11 feature detector‑matcher pairs on synthetic, lunar‑testbed, and terrestrial analogue data to evaluate front‑end matching for lunar VO/SLAM. SuperPoint with LightGlue is fastest, outperforms SIFT in rotation‑only tests, and maintains tracking across all lunar sequences, though SIFT excels in joint rotation‑translation accuracy on synthetic data.
DeXplicit: Making Physical State Explicit in Dexterous Manipulation World Models
AI summaryDeXplicit explicitly predicts metric 3D object motion and deformation before rendering, enabling structured future world states. It reduces moved-point error by 27‑40% and boosts object fidelity by 47%, while 57× faster than a 14B bi‑directional model.
WovenAnchor Matcher: Specialized Intra- and Inter-Image Context Modeling for Feature Matching
AI summaryWAM introduces role‑specialized context aggregation for semi‑dense matching: intra‑image propagation via WovenMamba and inter‑image retrieval via AnchorCrossAttention. On MegaDepth, WAM achieves 66.1 pose AUC@5°, 2.0 points above JamMa, demonstrating the benefit of the two complementary operators.
DRILL: Training World Models to Improve Policies, Not Predict Pixels
AI summaryDRILL trains world models to maximize downstream policy returns instead of pixel accuracy, using a bilevel update. The resulting DRILL‑IMG policy outperforms existing methods by 13.1 points in success rate across multiple VLA benchmarks.
Loop to Look Better: Learning Adaptive Depth for Question-Guided Perception in Vision-Language Models
AI summaryLoopLook adds a lightweight, question-guided recurrence to VLM visual encoders, iteratively refining mid-to-late layers until a learned stop condition. It outperforms baselines by up to 8.6% on MMStar, 8.1% on HallusionBench while adding only ~5‑6% TTFT.
One Evaluation Is Enough: Inversion-Free Image Editing with Trajectory Correction
AI summaryThe paper introduces Endpoint-Calibrated Transport (ECT), an inversion-free image editing method requiring only a single network evaluation. ECT achieves high-quality edits in one step by combining local tangent integration with a constrained endpoint correction.
RECAP: Looking Once Is Not Enough for Vision-Language Reasoning
AI summaryRECAP introduces “visual revisit” to detect when vision‑language models drift from image evidence during chain‑of‑thought reasoning, and uses these revisit traces as a credit‑assignment signal. The method improves VLM accuracy by 6–6.5 points on HallusionBench without extra rewards or inference changes.
CAVC: Convergence-Adaptive Audio-Visual Cache for Modality-Asymmetric Diffusion Generation
Diffusion-based audio-visual generation methods typically perform joint denoising of audio and video latents throughout the entire sampling trajectory, repeatedly executing modality-specific denoising branches and cross-modal attention at every timestep. However, we observe that…
Direct Forcing: A Single-Stage Training Recipe for Few-Step Autoregressive Text-to-Video Generation
AI summaryDirect Forcing trains a few‑step autoregressive video model in one stage by jointly aligning reference videos with generated rollouts (causal continuation alignment) and matching their representation distributions (rollout distribution matching) using dynamic gradient balancing. Starting from Wan2.1‑1.3B, it learns a 4‑step generator in 250 updates, achieving 84.63 VBench Total and 86.01 Quality.
Cross-Spectral Feature Fusion for Multispectral Document Binarization
AI summaryThe paper introduces a multi‑branch conditional GAN that fuses visible, NIR, and structural features at encoder depth 3 using softmax gating, guided by edge maps from VIS‑NIR pairs. The F3 + NIR edge‑guided model improves binarization, achieving F1 = 0.7916, p‑FM = 0.8044, and IoU = 0.6654 over the baseline.
Verus-Insight: Towards Explainable Artifact Detection in AI-Generated Videos
AI summaryVerus‑Insight proposes a three‑step artifact detection pipeline—inspection planning, evidence verification, and evidence aggregation—trained with human‑anchored supervision. It achieves state‑of‑the‑art fine‑grained artifact recovery and video‑level AI‑generated video detection on the new Verus‑Bench.
HimmBO: High-Dimensional Interactive Model-Merging Bayesian Optimization for Image Generation
AI summaryHimmBO scales preference‑driven merging of text‑to‑image adapters to large collections by first screening adapters via level‑set estimation on pre‑generated images, then refining merge coefficients in a low‑dimensional subspace. It outperforms baselines while using 90% fewer image generations.
When Annotation Patterns Are Learned: Skin-Tone Dependence in Dermatology Concepts
AI summaryThe paper audits the SkinCon dermatology concept dataset and finds strong skin‑tone bias: erythema is under‑annotated on dark skin (OR 0.17) while hyper‑ and hypopigmentation are over‑annotated. Models trained on these labels replicate the bias and fail to correct it.
Query-Time Calibration of CLIP Features for Open-Vocabulary 3D Segmentation
AI summaryLifting CLIP features into 3D Gaussian scenes causes directional concentration; the paper introduces query‑time calibration that removes a shared scene direction, whitens text prototypes, and applies soft clustering/CSLS, boosting mean IoU on seven methods and four datasets without retraining.
H-2DGS: Height-Field Gaussian Splatting for Sharp and Deformable Neurosurgical Surface Reconstruction
Neurosurgical navigation commonly presents magnetic resonance imaging (MRI) as 2D slices, making it difficult to relate the anatomy in those slices to the cortical surface observed through the surgical microscope. We seek a common 3D visualization that combines a photorealistic…
Hierarchical Fine-Grained Cross-Modal Alignment via Dynamic Granularity Discovery and Optimal Transport
Fine-grained image-text retrieval requires a model not only to assess global semantic consistency between an image and a text, but also to model visual evidence for local textual semantics, including objects, attributes, quantities, actions, and relations, so as to distinguish…
Watch and Learn: Teaching Visual In-Context Learning by Removing Textual Shortcuts
AI summaryMIME is a task‑agnostic training set that removes textual shortcuts, forcing VLMs to learn visual in‑context rules from images alone. Fine‑tuning on MIME boosts vICL accuracy by 8–18 points while preserving single‑image performance, and a DTAD variant yields comparable gains with more consistent visual attention.
Making MLLM Latents Matter: Factor-Aware Predictive Learning for Cinematic Video Generation
Cinematic video generation requires both temporal inference from limited inputs (i.e., sparse rather than frame-by-frame guidance) and factor-aware control of character, scene, action, and camera. Multimodal large language models (MLLMs) offer useful priors for both…
ICGF-CXRMate: Image-Conditioned Clinical Context Compression and Gated Fusion for Chest X-ray Report Generation
AI summaryICGF‑CXRMate compresses patient data into fixed clinical tokens conditioned on the X‑ray, then fuses them via gated cross‑attention. It boosts RadGraph‑F1 to 28.86 and BLEU‑4 to 7.96 while cutting prompt length by 22.9 %.
Vision Transformers for Estimating Irradiance Using Data Scarce Sky Images
Accurate estimation of diffuse horizontal irradiance (DHI) is critical for optimising photovoltaic system performance and energy forecasting yet remains challenging in regions lacking comprehensive ground-based instrumentation. Recent advancements using Vision Transformers…
GateFormer: Forecast-Space Decomposition for Intra-Hour Solar Irradiance Prediction from Sky Images
Sub-hourly dispatch, balancing, and storage control increasingly depend on forecasts that anticipate rapid variability in solar generation. Accurate sky-image-based irradiance forecasting requires a combination of visual cloud dynamics with recent irradiance state while adapting…
Attention-guided classification and tumor localization of combined hepatocellular-cholangiocarcinoma
Combined hepatocellular-cholangiocarcinomas (cHCC-CCA) are a type of cancer that share components of two other primary liver cancers: hepatocellular carcinoma (HCC) and cholangiocarcinoma (CCA). Accurate diagnosis of cHCC-CCA, of which the behavior may resemble that of either…
Bound-Conditioned Latent Inference for Progressive Image Compression
AI summaryBound-Conditioned Latent Inference (BLI) reformulates progressive trit-plane coding as sequential latent inference with shrinking interval constraints, refining Gaussian parameters and latent estimates at each step. BLI achieves 17.78% BD-rate reduction over DPICT and 7.96% over CTC on Kodak, using 5.3× fewer parameters and lower latency.
GraphCAD: Geometry-Aware Diffusion on Edge-Adjacency Graphs for CAD Generation
Generative modeling for Computer-Aided Design (CAD) remains fundamentally challenged by the need to capture the intricate interplay between discrete topological connectivity and precise spatial geometry inherent in Boundary Representations (B-reps). Existing generative…
Diagnostic Loss: Asymmetric Supervision for Radiology Report Generation
Automated radiology report generation requires models to do more than produce fluent clinical narratives: they must explicitly state supported diagnoses while avoiding unsupported ones. However, existing medical vision-language models (MedVLMs) are typically trained with…
Revisiting Distribution Matching for State-to-Vision Policy Distillation in Laparoscopic Robotic Manipulation
AI summaryThis paper proposes a state-to-vision policy distillation framework for laparoscopic robotic manipulation, using a privileged state-based teacher to train a vision-based student. Experiments in simulated surgical environments show that this approach achieves more consistent and stable performance than direct reinforcement learning or imitation learning.
ResGVC: Residual Coding for High-fidelity Generative Video Compression
Generative video compression (GVC) methods usually employ decoder-side diffusion models to denoise reconstructed signals based on generative priors. Besides, most of these methods perform denoising in latent space rather than directly in pixel domain. In this paper, we propose…
A General-Purpose Dual-Path Interaction Framework for Multi-View Medical Image Segmentation: Alleviating Data Volume Dependency in Small-Sample Regimes
The performance of medical image segmentation depends on a large amount of annotated data, and most segmentation models experience a significant drop in performance in low-sample-size scenarios. Existing semi-supervised and weakly supervised methods rely on specific network…
I$^2$-KAN: Interpretable Intra- and Inter-Modal Multi-Pathway Representation Learning for Multimodal Sentiment Analysis
Multimodal sentiment analysis (MSA) aims to infer sentiment by integrating linguistic, acoustic, and visual information. However, many fusion methods integrate modality-specific representations and cross-modal interactions into joint representations, making it difficult to…
DRAFT: Decoding and Reconstructing 3D Room Layouts from fMRI and Camera Trajectories
While functional magnetic resonance imaging (fMRI) decoders reconstruct images and 3D objects, recovering the metric geometry of rooms from a moving viewpoint remains an open challenge. We introduce, to our knowledge for the first time, the task of reconstructing 3D layouts of…
Where Do Long Captions Fail? Position-Aware Diagnosis and Reinforcement Learning for Detailed Image Captioning
Detailed image captioning is usually framed as an amount problem: a model should mention more visual facts while avoiding hallucination. However, autoregressive VLMs generate captions as ordered sequences, and later generations may be produced under weaker or noisier visual…
Through a PRISM, Clearly: Step-Sequence Reasoning for Mistake Recognition in Procedural Videos
Procedural mistake recognition requires identifying deviations from valid multi-step activities. We study post-completion recognition from videos with benchmark-provided step boundaries. We introduce PRISM (Procedural mistake Recognition via Interpretable Semantic Modeling),…
WildDepth-v2: A Synchronized RGB--LiDAR--Radar Dataset and Benchmark for Metric Depth, Motion, and Tracking of Wildlife
Measuring animal distance and motion in metric units underpins abundance estimation, morphometrics, and movement ecology, yet monocular video lacks metric scale and wildlife datasets lack metric ground truth. We present WildDepth 2, extending the synchronized RGB-LiDAR WildDepth…
Pos4CHD: A Positive-Free Mask-Guided Counterfactual Synthesis Framework for Fetal Congenital Heart Defect Analysis
The scarcity of positive cases remains a major bottleneck in developing automated diagnostic systems for fetal Congenital Heart Defects~(CHDs), particularly for rare subtypes. Although normal cases are accessible with effort, positive cases are hard to collect due to low…
GM-GReFEL: Geometry-Guided Spatiotemporal State-Space Learning for Robust Dynamic Facial Expression Recognition
Recognizing facial expressions in unconstrained video suffers from two coupled problems: self-attention grows quadratically in cost with clip length, and crowdsourced labels are noisy. We present GM-GReFEL, a spatiotemporal state-space network with a geometry-aware decision…
RankAlign: Vision-Language Representation Alignment via Rank Transformation
While the Platonic Representation Hypothesis posits that vision and language models converge toward compatible latent geometries, recovering class-level correspondences between unpaired modalities without cross-modal supervision remains challenging. Existing relational matching…
Navigating human-robot interaction challenges in private and residential aged care homes
This chapter explores how robotic systems and ambient intelligence can enhance care for older adults in residential aged care (RAC) homes and private homes. Although these settings differ, we frame both as extreme and challenging environments for human-robot interaction (HRI),…
WildFakeToolBench: Benchmarking AI-Generated Image Detectors with Web-Sourced Image Processing
As AI generated images are getting popular these days, discussions on such generations are widely existed in current public forums, software repositories, and photography tutorials. Moreover, with the help of LLM-assisted code translation, malicious actors can adopt such…
Beyond Watching and Searching: Benchmarking Agents for Video-centered Deep Research Reports
Deep research agents synthesize open-web evidence into long reports, but current benchmarks do not evaluate research centered on a source video. We introduce Video-Centered Deep Research Report Generation. Given a video and user request, an agent identifies research questions,…
HiSTEVE 1.0: Harnessing Human-Interface VLA Agents for Open-World Games
Building virtual embodied agents that operate in open-ended games like human players is a long-standing goal requiring both long-horizon autonomy and reliable native-interface control. As a procedurally generated open world, Minecraft provides a natural testbed for studying the…
Memory before Goal: Role of Pre-History Memory for Humanoid Loco-Manipulation
Humanoid loco-manipulation in dynamic scenes is non-Markovian: cues vanish as objects change or are hidden, or leave the view as the robot turns and walks, yet whole-body vision-language-action (VLA) policies see only the present. We present WholeBodyMem, an object-guided…
RaSFNet: Ray-aligned and Scan-aware Fusion Network for LiDAR Semantic Segmentation
LiDAR point cloud semantic segmentation is a fundamental component of 3D perception for autonomous driving and robotics. Projection-based methods map 3D point clouds into 2D representations for feature extraction using 2D networks. However, aggregating multiple points during…
Inside-JEPA: From Prediction to Relational Grounding in Latent World Models
Joint-embedding predictive architectures (JEPAs) learn world models in latent space, enabling efficient planning but leaving their states opaque. We introduce Inside-JEPA, which places relational channels for ordered entity pairs inside the JEPA latent state and shows that an…
A Visual Grounding Benchmark for Retinal OCT with Multimodal Large Language Models
Multimodal large language models (MLLMs) can answer clinical questions about medical images, but trustworthy use requires visual grounding: models should localize the pathology they report rather than rely on spurious shortcuts. Whether MLLMs can visually ground clinical…
Moving Regions, Not Moving Objects: Segmenting What Actually Moves under Camera Motion
Moving object segmentation is conventionally formulated at the sequence and object level, marking an entire instance as moving once it moves anywhere in a video. Recent instantaneous formulations distinguish moving from stationary intervals, but still generate masks at the…
GLOW-FM: Efficient Generative Low-Light Image Enhancement with One-Step Latent Flow Matching
Low-light image enhancement (LLIE) aims to recover well-exposed and perceptually faithful images from severely degraded observations. Recent diffusion-based generative methods achieve strong perceptual quality by leveraging powerful image priors, but their iterative sampling…
Lightweight Frame Compensation for Real-Time Polyp Segmentation under Frame Skipping
Real-time polyp segmentation in computer-assisted colonoscopy is hampered by the difficulty of deploying high-accuracy models under strict latency budgets. Frame skipping, which performs inference every K frames and reuses or interpolates the rest, improves throughput but…
Reinforcing intra-operative tumour localisation with weak supervision from pre-operative appearances
Limited tumour visibility on intra-operative images complicates surgical navigation, including liver hepatocellular carcinoma (HCC) targeting on computed tomography (CT) imaging. This makes labelled datasets difficult to curate for deep learning, and transferring pre-operative…
ReCoG: Relational Concept Graph for Training-Free Personalization
Training-free personalization adapts a frozen vision-language model (VLM) to recognize user-defined concepts from only a few exemplars by retrieving concept entries at inference time. While recent retriever-based methods enrich each concept with descriptive evidence, they…
Retrieval-Augmented Online Textual Memory for Episodic Memory Video Question Answering
Long egocentric videos pose a central challenge for multimodal language models: the visual stream must be compressed into a memory representation that can be searched efficiently while still preserving the evidence needed for question answering. Building on an online…
RecallGS: Geometry-Grounded Recall for Online Semantic Gaussian Splatting
Online semantic 3D Gaussian Splatting (3DGS) enables open-vocabulary scene understanding from streaming observations. However, existing methods recover geometric context missing from image-derived semantic features using an additionally learned 3D network, incurring extra…
PiHand: Latent Hand-Object Interaction Priors for Physics-Based Policy Learning
Synthesizing physically grounded dexterous hand-object interactions is important for VR/AR, gaming, and robotics, yet remains challenging because subtle contact dynamics, diverse object geometries, and target trajectories must be jointly respected while maintaining natural…
Data-Driven Vision-Language Models for Remote Sensing: A Survey
During the deep learning era, innovations in remote sensing (RS) vision models primarily focused on optimizing network architectures for specific tasks and conducting end-to-end training. However, with the advent of the Vision-Language Models (VLMs) era, constrained by…
Stream4D: Streaming 4D Reconstruction with Training-free Motion Decoupling via Causal KV-Cache Attention
AI summaryStream4D removes pose drift in streaming 4D reconstruction by decoupling motion without training: it extracts motion cues from causal Transformer KV caches, fuses them with depth priors to create dynamic masks, and soft‑key masks them during attention, boosting pose accuracy and reducing ghosting.
MVLoMa: Multi-View Local Feature Matching with Any Geometry Foundation Model
Local feature matching is typically formulated for image pairs, although structure-from-motion and visual localization commonly match one reference image against many retrieved or covisible neighbours. Applying a pairwise matcher independently to these centre-neighbour edges…
WeVisOne-VL: Unified Visual Perception from Objects to Pixels
Existing vision-language models (VLMs) remain unreliable for fine-grained visual perception. In this report, we present WeVisOne-VL (WeChat Vision All-in-One), a powerful VLM that unifies object-level and pixel-level perception through structured next-token prediction. It…
A Task-Centric Survey of Long-Document Visual Question Answering: Taxonomy, Evolution, and Future Directions
Visual document understanding is essential for acquiring knowledge from visually rich documents and is attracting growing research interest. Existing surveys emphasize architectures or technical approaches, with limited attention to how task bottlenecks shape technical choices…
A Sparse Topometric Graph for Scalable Cross-Session Indoor Navigation
A conventional embodied agent in an indoor environment builds a map by accumulating noisy real-world measurements and results from heavy inference to guide navigation. While the detailed measurements may differ due to scene changes or sensor settings, the unique topological…
MixPT: Point Transformer with Interleaved Window and Global Attention
Point cloud transformers aim to capture detailed local geometry and relationships across a scene at a manageable computational cost. Two common approaches are window attention and global attention. They differ in how they enable scene-wide interaction, which refers to…
AnyGameRenderer: Entity-Consistent Generative Rendering Across Games and an Entity-Level Benchmark
AI coding agents can now build small playable games in minutes, but the results usually look like programmer art: gray blockouts, colored geometric primitives, and flat semantic blobs. We present AnyGameRenderer, a video-to-video generative renderer that turns such footage into…
Ask About the Right Moment: Training-Free Object Identity Reasoning for Video-Language Models
Many questions about a video concern one physical object. Where does the ball end up after the cups are shuffled? Which card is the queen? How often was the mug covered? Video-language models (Video-LMs) recognise such objects in single frames but do not keep them apart over…
MoG-VLM: Efficient Video Anomaly Detection through Motion-Guided Vision-Language Models
Vision-Language Models (VLMs) have shown strong potential for video anomaly detection, but dense frame-level inference makes them impractical for real-time deployment. Existing acceleration commonly relies on uniform sampling, which ignores content dynamics and can either waste…
Self-Supervised Multibeam Sonar Restoration via Coherent Forward Modeling
AI summaryThis paper introduces a physics-consistent self-supervised framework for joint despeckling and deblurring of multibeam sonar using multi-frame sequences. Experiments on real datasets show it effectively suppresses speckle and preserves structures without requiring clean ground truth.
AgentVQA: A Unified Benchmark for Agentic Visual Understanding
Vision-language models (VLMs) can perform a broad range of tasks across diverse settings, yet their performance in agentic contexts remains poorly understood. Although online evaluations in simulators and real-world environments are essential for measuring interactive agent…
Evaluation of Uncertainty-Aware Multi-Software Ensembles for Hippocampal Segmentation
Accurate hippocampal segmentation can be a useful tool for diagnosing and monitoring neurological conditions such as Alzheimer's disease and epilepsy. While numerous automated segmentation methods exist, their clinical adoption remains limited. Reliable uncertainty assessment…
Diffusion-Based Synthesis of Complete 3D Biparametric Prostate MRI
Synthetic biparametric MRI (bpMRI) may help alleviate the limited availability of large and representative prostate MRI datasets for deep learning applications. In this work, we present a conditional 3D diffusion framework for anatomy-conditioned synthesis of complete prostate…
LACToS: Local Anomaly Detection via Conditional Typicality of Superpixels
AI summaryLACToS identifies anomalous image regions by measuring statistical typicality of superpixel groups, using an efficient proxy for typicality estimation. Across medical, face, and X-ray security benchmarks, it outperforms previous methods, especially when the inlier distribution is multimodal.
Squaring-the-Sphere: Distortion-Free and Seamless Panorama Generation via CubeAtlas
Generating 360-degree panoramas requires both local perspective fidelity and global spherical continuity. Existing representations face a trade-off: equirectangular projection (ERP) enables seamless panorama modeling but suffers from severe distortion and a domain gap to…
CellTrack-Bench: Microscopic Cell Tracking Benchmark for Vision-Language Models
Large language models have shown outstanding performance on text-first tasks such as coding, math, and writing. Recent efforts are shifting models' training to include scientific problems that often live in highly specialized domains beyond text, such as protein structure and…
Beyond Best-of-$k$: Prior-Consistent DAPS for Ambiguous Nonlinear Inverse Problems
Diffusion models have become strong priors for solving inverse problems. However, nonlinear inverse problems such as phase retrieval remain challenging when multiple measurement-consistent solutions are distinguishable only by the prior; we call such measurements ambiguous.…
F2G-Pose: Geometry-Aware Foundation Feature Lifting for Direct RGB-D Category-Level Object Pose Estimation
This paper studies category-level RGB-D object pose estimation, which recovers an object's 3D rotation and translation without instance-specific CAD models or reference views. This task is highly ambiguous due to intra-class variations, symmetries, and partial observations.…
HiP3R: Hierarchical Pose Graphs for Long-Horizon 3D Reconstruction
Feed-forward models accurately estimate camera motion and 3D geometry from short video windows. However, computational and memory costs prevent direct inference on sequences containing thousands of frames. Long videos must therefore be processed as overlapping chunks. This…
WorldForge V2: Eliciting Camera-Trajectory Alignment in Video Diffusion Models
AI summaryWorldForge V2 uses training‑free warped references and selective flow‑guided latent channels to refine video diffusion outputs along a target camera path, then trains lightweight Trajectory LoRA adapters with minimal data. The method delivers state‑of‑the‑art camera control and video quality.
UniVOE: A Removal-Driven Framework for Interaction-Aware Universal Video Object Editing
Video object removal and injection are reciprocal editing operations between object-present and object-absent video states, yet they differ substantially in training difficulty. Removal can be effectively bootstrapped from video inpainting and further adapted to erase…
GigaBench: Benchmarking Learnable Representations for Gigapixel Image Compression and Super-Resolution
Learnable Representations (LRs) can fit parametric families of functions to diverse kinds of signals. These emergent neural-native data formats are gathering interest in the scientific community due to their compactness, resolution-agnosticism and amenability to direct neural…
OmniRay: Camera-Agnostic Monocular 3D Reconstruction with Composite Ray Fields
AI summaryOmniRay predicts a ray direction and radial distance per pixel using a three‑stage ray field—monotone radial base, spherical‑harmonic residual, and free‑form residual—allowing camera‑agnostic monocular 3D reconstruction. It outperforms prior methods on diverse camera types.
Gate3DGen: A Pre-processing Model that Selectively Purifies Real-World Images for 3D Generation
Single-image 3D generation has achieved high fidelity on curated benchmarks, yet degrades sharply on real-world images with uneven illumination, blur, and occlusions. Purifying such inputs requires deciding when to intervene and preserving the visual cues needed for 3D…
BrickWright: Agentic Creation of Physically Feasible Brick Builds
AI summaryBrickWright gives coding agents tools to build physically valid LEGO models from text or image prompts, using a lightweight assembly API and a validator. Guided agents achieve 100% physical feasibility on 512 prompts, surpassing baseline and specialized models.
SpectralCanary: Embarrassingly Simple Low-Frequency Traces for T2I Dataset-Use Auditing
AI summarySpectralCanary embeds a low‑frequency mask into every image of a dataset to leave a detectable audit trace in T2I models without altering captions or requiring per‑image tuning. It achieves ~91% per‑generation TPR@$1\%$ FPR and ~95% model‑level accuracy while preserving image fidelity across diverse domains.
Reading, Writing, Sharing, and Interacting with the World through Video
AI summarySurvey unifies video foundation models and video world models under the lens of latent temporal world dynamics, categorizing work into reading, writing, sharing, and interacting. It highlights a shift from clip-level tasks to persistent, controllable, causally valid world states and interaction-based evaluation.
Sensorimotor Alignment: A Platonic Proxy for Embodied Policy Success
AI summarySensorimotor Alignment (S²) is a training‑free offline metric that quantifies how well a model’s sensory representation matches the geometry of physical motion. The paper shows S² strongly correlates with downstream robot task success and can identify promising vision‑language backbones before costly fine‑tuning.
A Unified Framework for Image-to-3D Part Generation via Variable Granularity
AI summaryFlexPart unifies image‑to‑3D part generation with a single model that accepts variable‑granularity conditions (points, boxes, masks) and uses gated adaptive modulation and asymmetric guidance to align representations. On PartVerseXL, it achieves 16.6 % lower Part CD than TripoSG + SAM3D and 10.5 % lower than OmniPart, while keeping whole‑mesh quality competitive.