vision-language model
PulseAugur coverage of vision-language model — every cluster mentioning vision-language model across labs, papers, and developer communities, ranked by signal.
- 2026-05-19 research_milestone A new method is proposed to improve out-of-distribution visual document understanding in VLMs. 来源
17 天有情绪数据
-
GPT-5 Mini leads Agentick benchmark, but no agent paradigm dominates
The new Agentick benchmark, which assesses various AI agents across 37 tasks, shows GPT-5 Mini achieving the top score of 0.309. However, no single agent paradigm, including reinforcement learning, LLM, VLM, or hybrid a…
-
New SAEgis framework detects adversarial attacks on vision-language models
Researchers have developed a new framework called SAEgis to detect adversarial attacks on vision-language models (VLMs). This method utilizes sparse autoencoders (SAEs) as a plug-and-play module, requiring no additional…
-
ChartZero uses synthetic data to extract chart data without real-world annotation
Researchers have developed ChartZero, a novel framework designed to extract data from line charts with zero-shot capabilities. This approach bypasses the need for real-world annotations by training exclusively on synthe…
-
CompART training improves VLM multi-object grounding and visual understanding
Researchers have developed a new training method called Compositional Attention-Regularized Training (CompART) to improve how Vision-Language Models (VLMs) handle complex, multi-object references. Current VLMs struggle …
-
GeoStack framework enables efficient VLM knowledge composition, preventing catastrophic forgetting.
Researchers have developed GeoStack, a novel framework designed to enhance knowledge composition in Vision-Language Models (VLMs). This approach addresses the issue of catastrophic forgetting, where models lose previous…
-
Consensus Entropy improves VLM OCR accuracy by measuring inter-model agreement
Researchers have developed a new metric called Consensus Entropy (CE) to assess the reliability of Optical Character Recognition (OCR) outputs from Vision-Language Models (VLMs). CE measures the agreement between multip…
-
Researchers propose new framework for generative recommendation systems
Researchers have developed a new framework to improve the generation of Semantic IDs (SIDs) for generative recommendation systems. This approach addresses issues of information and semantic degradation by integrating de…
-
PhysForge generates physics-grounded 3D assets for virtual worlds and embodied AI
Researchers have introduced PhysForge, a novel framework designed to generate physics-grounded 3D assets for interactive virtual worlds and embodied AI. This system addresses the limitations of existing methods by focus…
-
New AI models InterMesh and Anny-Fit advance 3D human pose and shape recovery
Researchers have developed InterMesh, a new framework for multi-person human mesh recovery that explicitly incorporates human-environment interaction information. This approach enhances pose and shape estimation by enri…
-
VLM pipeline enables viewpoint-agnostic grasping for robots with partial observations
Researchers have developed a new end-to-end pipeline for language-guided grasping that enhances the robustness of mobile manipulators in cluttered environments. This system uses visual-language models (VLMs) and partial…
-
Researchers unveil new stealthy backdoor attacks on AI models using diffusion and style features
Researchers have developed new methods for backdoor attacks on advanced AI models, specifically targeting Vision-Language Models (VLMs) and Diffusion Models (DMs). One approach, CBV, uses diffusion models to create natu…
-
New GLANCE framework enhances VLM agents with curiosity-driven visual-linguistic exploration
Researchers have developed a new framework called GLANCE to enhance the exploration capabilities of Visual-Linguistic Model (VLM) agents. This framework aims to improve how these agents navigate complex and partially ob…
-
New benchmark reveals video models forget long-term context
Researchers have introduced SceneBench, a new benchmark designed to evaluate video understanding models' ability to retain context over long videos, particularly across different scenes. Their findings indicate that cur…
-
VISTA benchmark launched for advanced VLM spatio-temporal interaction analysis
Researchers have introduced VISTA, a new benchmark designed to evaluate the spatio-temporal understanding capabilities of Vision-Language Models (VLMs). Unlike existing benchmarks that focus on simple actions and limite…
-
Researchers propose Gromov-Wasserstein distance for VLM vision encoder selection
Researchers have developed a new method for selecting optimal vision encoders for Vision-Language Models (VLMs). Traditional approaches, like choosing encoders with high accuracy or large size, were found to be ineffect…
-
Chain of Evidence framework enables pixel-level visual attribution for retrieval-augmented generation
Researchers have developed a new framework called Chain of Evidence (CoE) to improve iterative retrieval-augmented generation (iRAG) systems. CoE utilizes Vision-Language Models to directly analyze screenshots of retrie…
-
Coral and CoRAL systems optimize LLM serving and robotic control
Researchers have developed two distinct systems named Coral and CoRAL. Coral is an adaptive system designed for cost-efficient serving of multiple large language models across heterogeneous cloud GPUs, aiming to optimiz…
-
Robots gain semantic understanding with VLM and adaptive memory
Researchers have developed a "Semantic Autonomy Stack" to enable indoor mobile robots to understand natural language instructions, overcoming the latency and memory limitations of current Vision-Language Models (VLMs). …
-
GeoThinker framework actively integrates geometry for advanced spatial reasoning
Researchers have developed GeoThinker, a novel framework that enhances spatial reasoning in multimodal large language models (MLLMs) by actively integrating geometric information. Unlike previous passive fusion methods,…
-
New benchmarks tackle 'Entity Identity Confusion' in LLM knowledge editing
Researchers have identified a new failure mode in multimodal knowledge editing called Entity Identity Confusion (EIC), where edited vision-language models incorrectly associate new entity information with original image…