Every paper checked — nothing hidden

All papers

Every paper checked for brief #4 of 12–19 Oct 2026 — nothing is hidden. Brief papers first.

All17,313 In brief24 More (on-topic)3,333 Not relevant13,754 Not assessed87 Older352 Could not confirm93 Blogs with no paper266 All blogs284

Paper	Topic	Relevance
Adaptive Observation Control for Active Visual Recognition Adapts both field of view and fixation counts per image to optimize computation in active visual recognition. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant General active visual recognition rather than manufacturing defect inspection.
Self-Conditioning Enables Deep Sequential Reasoning in Flow Matching Proves theoretically that self-conditioning enables genuine multi-step reasoning in flow matching where standard samplers fail. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Theoretical study of flow matching, not industrial defect detection.
Self-Conditioning Enables Deep Sequential Reasoning in Flow Matching Proves theoretically that self-conditioning enables genuine multi-step reasoning in flow matching where standard samplers fail. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Does not address deployment safety, hallucination, or guardrails.
Self-Conditioning Enables Deep Sequential Reasoning in Flow Matching Proves theoretically that self-conditioning enables genuine multi-step reasoning in flow matching where standard samplers fail. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Investigates flow matching expressivity rather than LLM benchmarks or evaluation methods.
Self-Conditioning Enables Deep Sequential Reasoning in Flow Matching Proves theoretically that self-conditioning enables genuine multi-step reasoning in flow matching where standard samplers fail. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Addresses circuit complexity and reasoning depth rather than model efficiency or compression.
Mapping Arctic Polynyas from Sea Ice Concentration Data Using Hierarchical Partitioning and Differentiated Region Growing: Sensitivity to Threshold and Product Choices Delineates Arctic coastal polynyas from satellite sea ice concentration data using hierarchical partitioning and region growing. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Applies classical image segmentation algorithms to sea ice data without multimodal models.
Mapping Arctic Polynyas from Sea Ice Concentration Data Using Hierarchical Partitioning and Differentiated Region Growing: Sensitivity to Threshold and Product Choices Delineates Arctic coastal polynyas from satellite sea ice concentration data using hierarchical partitioning and region growing. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Focuses on satellite sea ice mapping, not manufacturing or industrial vision.
SPECTRA-VLA: Base-Fine Hyperspectral Token Transfer at Frozen Vision-Language-Action Interfaces Transfers hyperspectral tokens into frozen vision-language-action models via learned continuous residuals without retraining. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Focuses on robotics VLA adaptation rather than industrial visual inspection.
Remember, Revisit, Revise: Scaffold Memory for Long-Horizon 3D Geometry Equips frozen 3D geometry foundation models with an associative external memory to revise past predictions across long horizons. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Deals with 3D camera pose and geometry, not industrial defect detection.
Remember, Revisit, Revise: Scaffold Memory for Long-Horizon 3D Geometry Equips frozen 3D geometry foundation models with an associative external memory to revise past predictions across long horizons. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Centers on geometric computer vision rather than multimodal learning.
MicroSDF: Microfacet-Driven Hybrid Neural SDFs for Mixed-Reflectance Surface Reconstruction Reconstructs mixed-reflectance 3D surfaces by integrating microfacet normal consistency into neural implicit distance fields. Peer-reviewed	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Focuses on 3D reconstruction graphics, not industrial inspection.
MicroSDF: Microfacet-Driven Hybrid Neural SDFs for Mixed-Reflectance Surface Reconstruction Reconstructs mixed-reflectance 3D surfaces by integrating microfacet normal consistency into neural implicit distance fields. Peer-reviewed	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Addresses 3D graphics rendering rather than cross-modal understanding.
Enhancing Latent World Models with Associative Memory Adds content-addressable associative memory to latent world models to prevent collapse and improve long-horizon planning. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Focuses on latent planning benchmarks, not industrial visual inspection.
Enhancing Latent World Models with Associative Memory Adds content-addressable associative memory to latent world models to prevent collapse and improve long-horizon planning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on embodied planning dynamics, not AI deployment safety or guardrails.
Enhancing Latent World Models with Associative Memory Adds content-addressable associative memory to latent world models to prevent collapse and improve long-horizon planning. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Addresses reinforcement learning dynamics, not vision-language multimodal models.
Enhancing Latent World Models with Associative Memory Adds content-addressable associative memory to latent world models to prevent collapse and improve long-horizon planning. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Presents a novel latent model, not an LLM benchmark or evaluation study.
Generative World Simulation with DAO and JING Couples a persistent symbolic world engine with an egocentric generative model for continuous interactive audiovisual simulation. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Virtual environment simulation, not industrial computer vision.
A Diagnostic Study of World-Space Human Recovery Under High Camera Dynamics Identifies scale mismatches as the cause of trajectory drift in monocular human recovery under rapid camera movement. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Human pose recovery from dance videos, not industrial inspection.
A Diagnostic Study of World-Space Human Recovery Under High Camera Dynamics Identifies scale mismatches as the cause of trajectory drift in monocular human recovery under rapid camera movement. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Deals with monocular human mesh recovery, not multimodal learning.
Object Pose from 2-D to 3-D Point and Line Correspondences Presents a trust-region optimization method using dual quaternions to estimate 3D object pose from 2D correspondences. Peer-reviewed	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Geometric camera pose estimation, not industrial defect detection.
Object Pose from 2-D to 3-D Point and Line Correspondences Presents a trust-region optimization method using dual quaternions to estimate 3D object pose from 2D correspondences. Peer-reviewed	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Standard geometric 3D vision, not multimodal models.
HyperLoop: Hyperspectral-Assisted Loop Closure Detection in 3D-SLAM for Planetary Exploration Incorporates hyperspectral imaging descriptors into 3D Gaussian Splatting loop closure to prevent visual aliasing in planetary SLAM. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Planetary rover SLAM navigation, not industrial manufacturing quality control.
HyperLoop: Hyperspectral-Assisted Loop Closure Detection in 3D-SLAM for Planetary Exploration Incorporates hyperspectral imaging descriptors into 3D Gaussian Splatting loop closure to prevent visual aliasing in planetary SLAM. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Hyperspectral robotics SLAM, not multimodal vision-language architectures.
Discriminative Score Function: Turning Pretrained Models into Functional Generative Priors Extracts generative image priors directly from the loss-gradient geometry of pretrained discriminative models without retraining. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Mentions calibration secondary use, but main focus is generative score functions.
Discriminative Score Function: Turning Pretrained Models into Functional Generative Priors Extracts generative image priors directly from the loss-gradient geometry of pretrained discriminative models without retraining. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant General generative score extraction from classifiers, not industrial vision.
Discriminative Score Function: Turning Pretrained Models into Functional Generative Priors Extracts generative image priors directly from the loss-gradient geometry of pretrained discriminative models without retraining. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Method for generative priors, not LLM benchmark evaluation.
Discriminative Score Function: Turning Pretrained Models into Functional Generative Priors Extracts generative image priors directly from the loss-gradient geometry of pretrained discriminative models without retraining. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on image generation priors, not AI safety or deployment guardrails.
Discriminative Score Function: Turning Pretrained Models into Functional Generative Priors Extracts generative image priors directly from the loss-gradient geometry of pretrained discriminative models without retraining. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Focuses on image-space discriminative models, not multimodal models.
Physics-Shaped, Not Physical: Auditing Physics-Guided Underwater Image Restoration Audits physics-guided underwater image restoration models, showing image-formation priors provide minimal benefit under full training data. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Focuses on single-image physics restoration, not multimodal foundation models.
Physics-Shaped, Not Physical: Auditing Physics-Guided Underwater Image Restoration Audits physics-guided underwater image restoration models, showing image-formation priors provide minimal benefit under full training data. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Underwater image restoration, not industrial quality control or inspection.
Q-MAGIC: Quad-only Mesh Autoregressive Generation via Singularity-Aware Circular Relative Offset Tokenization Generates quad-only 3D meshes autoregressively using a singularity-aware tokenization scheme that preserves strict topology. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Focuses on 3D mesh serialization, not multimodal models.
Q-MAGIC: Quad-only Mesh Autoregressive Generation via Singularity-Aware Circular Relative Offset Tokenization Generates quad-only 3D meshes autoregressively using a singularity-aware tokenization scheme that preserves strict topology. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant 3D geometric mesh generation, not industrial visual inspection.
Computational compressive imaging reconstruction with a simulated-annealing-based Ising solver Formulates computational compressive imaging reconstruction as an Ising model solved via simulated annealing. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Optical compressive image reconstruction, not industrial visual defect detection.
Computational compressive imaging reconstruction with a simulated-annealing-based Ising solver Formulates computational compressive imaging reconstruction as an Ising model solved via simulated annealing. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Deals with optical physics reconstruction, not multimodal language-vision models.
FPV-CRASH: A Dataset for Offline Crash Moment Localization in Drone First-Person-View Videos Introduces a benchmark dataset to localize impact moments in drone first-person-view crash videos for post-flight analysis. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Addresses video crash localization, not multimodal reasoning architectures.
FPV-CRASH: A Dataset for Offline Crash Moment Localization in Drone First-Person-View Videos Introduces a benchmark dataset to localize impact moments in drone first-person-view crash videos for post-flight analysis. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Drone crash video localization, not factory or manufacturing inspection.
Improving Category-Aware Active Domain Adaptation via Uncertainty-Guided Region Coverage Improves active domain adaptation for difficult classes using region-based positive manifold coverage and uncertainty scoring. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Focuses on standard vision domain adaptation, not multimodal modeling.
Improving Category-Aware Active Domain Adaptation via Uncertainty-Guided Region Coverage Improves active domain adaptation for difficult classes using region-based positive manifold coverage and uncertainty scoring. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant General active domain adaptation, not industrial inspection applications.
PhyMetric: Diagnosing Physical Plausibility in Text-to-Video Generation via Scene-Level QA Introduces a diagnostic question-answering benchmark to evaluate whether text-to-video models follow real-world physical laws. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Evaluates text-to-video physical realism, not industrial defect detection.
What Matters for World Modeling in Vision-Language-Action Models? Evaluates four world-model integration paradigms to identify what improves vision-language-action policies in robotic tasks. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Robot manipulation world models, not industrial visual inspection.
VisibilitySplat: Ray-Traced Occlusion for Relighting with 3D Gaussian Splatting Combines 3D Gaussian Splatting with ray-traced occlusion estimation to enable relighting and material decomposition. Peer-reviewed	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Computer graphics inverse rendering, not multimodal text-image intelligence.
VisibilitySplat: Ray-Traced Occlusion for Relighting with 3D Gaussian Splatting Combines 3D Gaussian Splatting with ray-traced occlusion estimation to enable relighting and material decomposition. Peer-reviewed	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Novel view synthesis and relighting, not industrial vision inspection.
OmniPoser: Flexible Human Motion Recovery in the Wild with Masked Flow Matching Recovers full-body human motion from flexible sensor subsets using a cross-modal masking mechanism and single-step flow matching. Peer-reviewed	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Human pose recovery from wearables, not industrial visual quality control.
OmniPoser: Flexible Human Motion Recovery in the Wild with Masked Flow Matching Recovers full-body human motion from flexible sensor subsets using a cross-modal masking mechanism and single-step flow matching. Peer-reviewed	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Focuses on wearable sensor fusion and 3D motion, not multimodal models.
CoDy: Codec Dynamics as Complementary Priors for Unsupervised Temporal Action Localization Integrates video codec dynamics with visual features to improve unsupervised temporal action localization without manual annotations. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Video temporal action localization, not manufacturing or defect inspection.
CoDy: Codec Dynamics as Complementary Priors for Unsupervised Temporal Action Localization Integrates video codec dynamics with visual features to improve unsupervised temporal action localization without manual annotations. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Processes video codecs and RGB, not text-vision multimodal LLMs.
BluePixel: Toward Native 4K Agentic Image Editing Combines an adapted editing model with an agent harness to perform instruction-driven four-kilopixel image editing. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Focuses on 4K photo editing, not industrial manufacturing inspection.
On the Limits of Representation-Based Test-Time Adaptation for Multimodal LLMs Analyzes why feature-based test-time adaptation fails to recover accuracy on corrupted inputs in generative multimodal language models. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Investigates representation adaptation failure, not compression or efficient inference.
On the Limits of Representation-Based Test-Time Adaptation for Multimodal LLMs Analyzes why feature-based test-time adaptation fails to recover accuracy on corrupted inputs in generative multimodal language models. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Studies multimodal LLM adaptation, not industrial defect detection.
CARD: Closing the Teacher-Student Loop for Adversarial Robustness Distillation Develops a closed-loop distillation pipeline where student feedback refines teacher models to improve adversarial robustness. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Theoretical adversarial distillation method, not industrial computer vision.
CARD: Closing the Teacher-Student Loop for Adversarial Robustness Distillation Develops a closed-loop distillation pipeline where student feedback refines teacher models to improve adversarial robustness. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant General vision distillation, not cross-modal multimodal models.
CARD: Closing the Teacher-Student Loop for Adversarial Robustness Distillation Develops a closed-loop distillation pipeline where student feedback refines teacher models to improve adversarial robustness. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on training distillation, not LLM benchmark evaluation.
Quantifying Annotation Agreement in Object Detection with Optimal Transport Applies optimal transport metrics to quantify graded annotation agreement and label discrepancies in object detection datasets. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Focuses on 2D bounding box evaluation, not multimodal models.
Quantifying Annotation Agreement in Object Detection with Optimal Transport Applies optimal transport metrics to quantify graded annotation agreement and label discrepancies in object detection datasets. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant General object detection metric, not specific to industrial inspection.
SAM3RL: Memory Control via Reinforcement Learning for Visual Object Tracking Trains a reinforcement learning controller to manage frame memory dynamically for visual object tracking in foundation models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on memory management for tracking models, not LLM evaluation or benchmarks.
SAM3RL: Memory Control via Reinforcement Learning for Visual Object Tracking Trains a reinforcement learning controller to manage frame memory dynamically for visual object tracking in foundation models. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Addresses visual object tracking in generic videos, not industrial inspection or defect detection.
SAM3RL: Memory Control via Reinforcement Learning for Visual Object Tracking Trains a reinforcement learning controller to manage frame memory dynamically for visual object tracking in foundation models. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Addresses visual object tracking and video segmentation, not multimodal language-vision reasoning.
SAM3RL: Memory Control via Reinforcement Learning for Visual Object Tracking Trains a reinforcement learning controller to manage frame memory dynamically for visual object tracking in foundation models. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Studies visual memory controllers in tracking models, not language model compression or inference efficiency.
SAM3RL: Memory Control via Reinforcement Learning for Visual Object Tracking Trains a reinforcement learning controller to manage frame memory dynamically for visual object tracking in foundation models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Deals with reinforcement learning for video tracking memory, not deployment safety or guardrails.
Instruction Mining From Images: Constructing a Synthetic Dataset for Multimodal Learning ILSP DCVLR Competition Submission Generates a filtered synthetic dataset of image-based instruction-response pairs for training multimodal reasoning models. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Develops synthetic instruction datasets for multimodal models, not industrial visual inspection.

2,941–3,000 of 13,754← Previous Next →