Every paper checked — nothing hidden

All papers

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

All17,312 In brief24 More (on-topic)3,333 Not relevant13,753 Not assessed87 Older351 Could not confirm93 Blogs with no paper266 All blogs284

Paper	Topic	Relevance
ReState: Noise-Level-Aware State Correction for Training-Free Object Removal in Few-Step Diffusion Models Presents a training-free state-correction method to eliminate object residuals during object removal in few-step diffusion models. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Addresses diffusion inpainting and editing, not multimodal reasoning models.
ReState: Noise-Level-Aware State Correction for Training-Free Object Removal in Few-Step Diffusion Models Presents a training-free state-correction method to eliminate object residuals during object removal in few-step diffusion models. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on generative object removal in diffusion models, not industrial inspection.
GeoPano: Towards Geometrically Accurate 3D Reconstruction from a Single Panorama Introduces a feed-forward framework with panoramic attention to recover geometrically consistent 3D point clouds from a single panorama. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Deals with panoramic 3D reconstruction, not industrial visual inspection.
GeoPano: Towards Geometrically Accurate 3D Reconstruction from a Single Panorama Introduces a feed-forward framework with panoramic attention to recover geometrically consistent 3D point clouds from a single panorama. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Focuses on geometric single-panorama reconstruction, not multimodal vision-language systems.
Dusting the Phantom: Gaussian Allocation for Fading Dynamic Content Improves dynamic 3D Gaussian splatting for fast-moving objects by injecting seeds in under-reconstructed dynamic areas. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Focuses on 3D Gaussian volumetric video rendering, not multimodal models.
Dusting the Phantom: Gaussian Allocation for Fading Dynamic Content Improves dynamic 3D Gaussian splatting for fast-moving objects by injecting seeds in under-reconstructed dynamic areas. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Studies dynamic 3D Gaussian splatting, not industrial defect detection.
VLA-$Δ$: Learning Physical State Transition for Vision-Language-Action Models Proposes explicit learning of physical state transitions between current and future scenes to train robust vision-language-action robotic policies. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on robot manipulation benchmarks, not industrial quality inspection.
CoViG: Coarsened Vision GNN for Image Classification Reduces the inference computational cost of vision graph neural networks by applying graph coarsening via MinCUT pooling. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Tests general image classification on ImageNet, not industrial inspection.
CoViG: Coarsened Vision GNN for Image Classification Reduces the inference computational cost of vision graph neural networks by applying graph coarsening via MinCUT pooling. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Focuses on Vision Graph Neural Networks, not multimodal vision-language models.
AdaLoop: Thinking with the Vision Encoder Introduces a framework that enables vision-language models to iteratively adapt frozen visual encoder computation based on the input question. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Investigates VLM vision encoding architectures, not industrial defect detection.
CertEx: Exact Visual Retrieval at Scale by Certified Exclusion Presents a GPU nearest-neighbor search method that uses half-precision certified exclusion to deliver exact visual retrieval efficiently. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Provides exact kNN vector search optimization rather than multimodal models.
CertEx: Exact Visual Retrieval at Scale by Certified Exclusion Presents a GPU nearest-neighbor search method that uses half-precision certified exclusion to deliver exact visual retrieval efficiently. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on general vector retrieval algorithms, not industrial inspection.
Semantic Latent Alignment for Flow Matching with Invertible Bi-Lipschitz Adapters Proposes invertible bi-Lipschitz adapters to align pretrained latent spaces with semantic representations for flow matching without retraining encoders. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on generative coordinate alignment, not AI deployment safety.
Semantic Latent Alignment for Flow Matching with Invertible Bi-Lipschitz Adapters Proposes invertible bi-Lipschitz adapters to align pretrained latent spaces with semantic representations for flow matching without retraining encoders. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on flow matching latent spaces, not industrial inspection.
Semantic Latent Alignment for Flow Matching with Invertible Bi-Lipschitz Adapters Proposes invertible bi-Lipschitz adapters to align pretrained latent spaces with semantic representations for flow matching without retraining encoders. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Adapts generative latent spaces, not efficient serving or quantization.
Semantic Latent Alignment for Flow Matching with Invertible Bi-Lipschitz Adapters Proposes invertible bi-Lipschitz adapters to align pretrained latent spaces with semantic representations for flow matching without retraining encoders. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Explores latent alignment for generation, not LLM benchmark evaluation.
Semantic Latent Alignment for Flow Matching with Invertible Bi-Lipschitz Adapters Proposes invertible bi-Lipschitz adapters to align pretrained latent spaces with semantic representations for flow matching without retraining encoders. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Covers flow matching latent alignment, not multimodal foundation models.
When Should a Low-Light Pipeline Generate? Auditable Risk Control for Selective Generative Enhancement Presents an auditable framework that uses risk budgeting to selectively deploy generative enhancers over base restoration models in low-light images. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on selective generative enhancement, not model compression or serving.
When Should a Low-Light Pipeline Generate? Auditable Risk Control for Selective Generative Enhancement Presents an auditable framework that uses risk budgeting to selectively deploy generative enhancers over base restoration models in low-light images. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on image enhancement routing rather than LLM benchmarks.
When Should a Low-Light Pipeline Generate? Auditable Risk Control for Selective Generative Enhancement Presents an auditable framework that uses risk budgeting to selectively deploy generative enhancers over base restoration models in low-light images. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Addresses low-light image routing, not deployment safety or jailbreak prevention.
When Should a Low-Light Pipeline Generate? Auditable Risk Control for Selective Generative Enhancement Presents an auditable framework that uses risk budgeting to selectively deploy generative enhancers over base restoration models in low-light images. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Deals with low-light image restoration, not vision-language integration.
When Should a Low-Light Pipeline Generate? Auditable Risk Control for Selective Generative Enhancement Presents an auditable framework that uses risk budgeting to selectively deploy generative enhancers over base restoration models in low-light images. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Addresses low-light image enhancement, not industrial visual inspection.
L2GSSL: Label-Efficient Local-to-Global Self-Supervised Learning for Volumetric Healthcare Data Introduces a label-efficient self-supervised learning framework that splits volumetric CT scans into local slice and global context representations. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Covers medical image pretraining, not AI safety or guardrails.
L2GSSL: Label-Efficient Local-to-Global Self-Supervised Learning for Volumetric Healthcare Data Introduces a label-efficient self-supervised learning framework that splits volumetric CT scans into local slice and global context representations. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Addresses volumetric medical image representation, not multimodal vision-language models.
L2GSSL: Label-Efficient Local-to-Global Self-Supervised Learning for Volumetric Healthcare Data Introduces a label-efficient self-supervised learning framework that splits volumetric CT scans into local slice and global context representations. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Proposes self-supervised pretraining for CT scans, not LLM evaluation.
L2GSSL: Label-Efficient Local-to-Global Self-Supervised Learning for Volumetric Healthcare Data Introduces a label-efficient self-supervised learning framework that splits volumetric CT scans into local slice and global context representations. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on medical CT pretraining, not industrial defect detection.
L2GSSL: Label-Efficient Local-to-Global Self-Supervised Learning for Volumetric Healthcare Data Introduces a label-efficient self-supervised learning framework that splits volumetric CT scans into local slice and global context representations. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on label-efficient CT pretraining rather than model compression or inference efficiency.
CoDeFont: One-Shot Font Generation via Multi-Scale Factor-Specific Representation Learning Proposes a script-agnostic one-shot font generation framework separating style and content representations across multiple spatial scales. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on glyph font generation, not industrial inspection.
CoDeFont: One-Shot Font Generation via Multi-Scale Factor-Specific Representation Learning Proposes a script-agnostic one-shot font generation framework separating style and content representations across multiple spatial scales. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Deals with one-shot glyph style transfer, not multimodal reasoning.
SATOR: Backpropagation-Free 3D Test-Time Adaptation via Sample-Adaptive Token Routing Presents a backpropagation-free test-time adaptation method that purges unreliable 3D point cloud tokens without requiring candidate size search. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Evaluates 3D point cloud corruptions, not industrial defect detection.
SATOR: Backpropagation-Free 3D Test-Time Adaptation via Sample-Adaptive Token Routing Presents a backpropagation-free test-time adaptation method that purges unreliable 3D point cloud tokens without requiring candidate size search. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Focuses on 3D point cloud adaptation, not multimodal vision-language systems.
Perception or Reasoning? Diagnosing and Tackling Reinforcement Learning Bottlenecks in VLMs Introduces an attribution metric to diagnose whether perceptual limits or reasoning gaps constrain reinforcement learning gains in vision-language models. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Studies RL in vision-language models, not industrial inspection.
CLAD: Keypoint Matching Using Global Line Correspondences Presents a keypoint descriptor method that leverages global line correspondences to improve feature matching in repetitive or textureless scenes. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Presents visual keypoint matching, not multimodal text-image foundation models.
CLAD: Keypoint Matching Using Global Line Correspondences Presents a keypoint descriptor method that leverages global line correspondences to improve feature matching in repetitive or textureless scenes. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Evaluates general SLAM keypoint matching, not industrial manufacturing inspection.
ManiFusion: Unlocking the High-Throughput Generation via Superposition in Manifold Space Proposes a framework that increases image diffusion throughput by fusing and processing multiple generation samples simultaneously in manifold space. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Addresses diffusion sampling acceleration, not safety, guardrails, or alignment.
ManiFusion: Unlocking the High-Throughput Generation via Superposition in Manifold Space Proposes a framework that increases image diffusion throughput by fusing and processing multiple generation samples simultaneously in manifold space. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Proposes diffusion generation throughput optimization, not industrial inspection.
ManiFusion: Unlocking the High-Throughput Generation via Superposition in Manifold Space Proposes a framework that increases image diffusion throughput by fusing and processing multiple generation samples simultaneously in manifold space. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Concerns diffusion image generation throughput, not vision-language multimodal modeling.
ManiFusion: Unlocking the High-Throughput Generation via Superposition in Manifold Space Proposes a framework that increases image diffusion throughput by fusing and processing multiple generation samples simultaneously in manifold space. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on generative throughput rather than language model evaluation benchmarks.
RLDF: A Region-Level Deepfake Dataset and Benchmark for Partial-Face Manipulation Detection Introduces a benchmark and dataset of over two million images to evaluate deepfake detection models on partial-face manipulations. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Focuses on face-swap deepfake detection rather than multimodal vision-language reasoning.
RLDF: A Region-Level Deepfake Dataset and Benchmark for Partial-Face Manipulation Detection Introduces a benchmark and dataset of over two million images to evaluate deepfake detection models on partial-face manipulations. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Addresses facial deepfake detection, not industrial defect detection.
PET Quantification-Preserving Registration for Longitudinal Whole-Body PSMA PET/CT Presents a multi-modal registration framework for longitudinal whole-body PET/CT that preserves metabolic tumor quantification metrics. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on medical PET/CT registration, not industrial manufacturing inspection.
PET Quantification-Preserving Registration for Longitudinal Whole-Body PSMA PET/CT Presents a multi-modal registration framework for longitudinal whole-body PET/CT that preserves metabolic tumor quantification metrics. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Covers medical image registration, not multimodal language-vision foundation models.
CarryOn: A Dual-Loop Harness for Personalized and Reliable Robot Assistance Proposes a dual-loop harness augmenting frozen robot policies with persistent user preferences and proactive safety assurance. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on assistive robotics, not industrial manufacturing quality control.
Cross-Prompt Inference Mitigates Rare-Class Suppression in Imbalanced VLM Finetuning Introduces cross-prompt inference to recover rare-class recall drops caused by class-imbalanced finetuning in vision-language models. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on VLM prompt fine-tuning, not industrial inspection or defect detection.
GRASE: Geometry-native Real-to-sim Agentic Scene Modeling from Video Presents an agentic framework that iteratively refines 3D scene geometry from multi-view video captures against standing visual observations. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Studies 3D scene modeling from video, not industrial inspection.
GRASE: Geometry-native Real-to-sim Agentic Scene Modeling from Video Presents an agentic framework that iteratively refines 3D scene geometry from multi-view video captures against standing visual observations. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Focuses on geometry-native real-to-sim 3D scene reconstruction from video.
FilterFusion: Prior-Guided Adaptive Filter Fusion for Efficient Video Denoising Combines lightweight learned guidance with classical filter banks for efficient, low-latency video denoising in communication pipelines. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on video communication denoising, not industrial quality inspection.
FilterFusion: Prior-Guided Adaptive Filter Fusion for Efficient Video Denoising Combines lightweight learned guidance with classical filter banks for efficient, low-latency video denoising in communication pipelines. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Presents a unimodal video denoising pipeline without multimodal integration.
Rethinking Classifier-Free Guidance for TI2I Flow Matching Identifies early-stage issues with classifier-free guidance in flow matching models and proposes a correction for controllable image editing. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on text-guided image editing, not industrial manufacturing inspection.
Rethinking Classifier-Free Guidance for TI2I Flow Matching Identifies early-stage issues with classifier-free guidance in flow matching models and proposes a correction for controllable image editing. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Studies flow matching mechanisms, not LLM benchmarks or evaluation methods.
Rethinking Classifier-Free Guidance for TI2I Flow Matching Identifies early-stage issues with classifier-free guidance in flow matching models and proposes a correction for controllable image editing. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on flow guidance mechanics rather than model compression.
Rethinking Classifier-Free Guidance for TI2I Flow Matching Identifies early-stage issues with classifier-free guidance in flow matching models and proposes a correction for controllable image editing. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Studies generative controllability in flow matching, not AI deployment safety.
Decoupled Safety Control: A Safety-Control Algorithm for Training-Free Safety Guidance Proposes a decoupled inference-time guidance algorithm to filter unsafe outputs in text-to-image diffusion models without altering weights. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Studies text-to-image safety guidance, not industrial visual inspection.
MYOPIA: Understanding the Limits of Tree-Based Reasoning in Vision-Language Models Introduces a benchmark evaluating decision tree reasoning in vision-language models, revealing accuracy drops compared to direct zero-shot prompting. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Evaluates decision tree reasoning in VLMs, not industrial inspection.
Hear, Localize, and Reason: Spatially Aware Scene Understanding for Audio-Visual LLMs Releases a spatial audio-visual benchmark and model architecture designed to identify, localize, and reason about sound sources in scenes. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Studies audio-visual spatial reasoning, not industrial defect detection.
A Decision-Aware Evaluation Protocol for Uncertainty in Medical Image Segmentation Introduces an evaluation protocol for medical image segmentation uncertainty metrics based on clinical deferral policies and fixed review budgets. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Addresses medical image segmentation uncertainty, not industrial quality control.
A Decision-Aware Evaluation Protocol for Uncertainty in Medical Image Segmentation Introduces an evaluation protocol for medical image segmentation uncertainty metrics based on clinical deferral policies and fixed review budgets. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Focuses on medical vision segmentation uncertainty without multimodal components.
Training-Free Target-Aware Refinement of 3D Gaussian Language Map Proposes a training-free closed-form method to refine target object retrieval in 3D Gaussian language maps while preserving open-vocabulary queries. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Studies 3D Gaussian language maps, not industrial defect detection.
Anchor3DGS: Feed-Forward 3D Gaussian Splatting with Compact Anchor-Based Representation Presents a feed-forward 3D Gaussian splatting method that uses entropy-guided 3D anchors instead of pixel-aligned predictions for efficient reconstruction. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Proposes feed-forward 3D reconstruction, not multimodal vision-language modeling.
Anchor3DGS: Feed-Forward 3D Gaussian Splatting with Compact Anchor-Based Representation Presents a feed-forward 3D Gaussian splatting method that uses entropy-guided 3D anchors instead of pixel-aligned predictions for efficient reconstruction. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on 3D Gaussian Splatting reconstruction, not manufacturing inspection.

661–720 of 13,753← Previous Next →