Vision--Language Models
PulseAugur coverage of Vision--Language Models — every cluster mentioning Vision--Language Models across labs, papers, and developer communities, ranked by signal.
17 day(s) with sentiment data
-
New C2-Routing method boosts VLM efficiency by 36%
Researchers have developed a new method called C$^2$-Routing to improve the efficiency of vision-language models (VLMs) during inference. This technique adaptively combines image cropping and compression to reduce compu…
-
PhysMoE enhances spatial grounding in embodied VLMs with novel tokenization
Researchers have developed PhysMoE, a novel approach for embodied Vision-Language Models (VLMs) designed to improve spatial grounding and reduce decoding latency. The system utilizes modality-decoupled experts for spati…
-
New BEV3SR Framework Enhances 3D Spatial Reasoning in Vision-Language Models
Researchers have introduced BEV3SR, a novel framework designed to enhance 3D spatial reasoning capabilities in Vision-Language Models (VLMs) when processing video data. This training-free and backbone-agnostic system tr…
-
Subliminal learning transfers preferences in vision-language models
Researchers have investigated the phenomenon of subliminal learning in vision-language models (VLMs), where a teacher model imparts hidden preferences to a student model through unrelated training data. Experiments invo…
-
Robots learn zero-shot navigation in cluttered environments with MessyNav
Researchers have introduced MessyNav, a novel zero-shot framework designed for robots to navigate complex, cluttered environments. This system leverages vision-language models (VLMs) to perform explicit affordance reaso…
-
New TACT-KV method compresses 3D medical VLM caches, boosting batch size 8x
Researchers have developed TACT-KV, a novel method for compressing the key-value (KV) cache in vision-language models (VLMs) that process 3D medical imaging data. Unlike previous methods that only consider 1D or 2D tran…
-
New ATHAAR dataset targets handwritten Arabic manuscript recognition
Researchers have introduced ATHAAR, a new dataset designed to improve handwritten text recognition for medieval Arabic scientific manuscripts. Current vision-language models struggle with these texts due to complex liga…
-
New ARMs agent achieves over 90% attack success on Claude 4 Sonnet
Researchers have developed ARMs, an adaptive red-teaming agent designed to systematically assess the safety vulnerabilities of vision-language models (VLMs). ARMs employs diverse, multi-step strategies to elicit harmful…
-
New dataset LF-VideoQA challenges AI models on long-form video understanding
Researchers have introduced LF-VideoQA, a new dataset designed to evaluate the long-form understanding capabilities of Vision--Language Models (VLMs) on narrative videos. This dataset focuses on theory-of-mind questions…
-
M-PFT method improves calibration for vision-language models
Researchers have introduced M-PFT, a new method for prompt-tuning vision-language models (VLMs) that addresses calibration issues. Prompt-tuned VLMs often exhibit underconfidence on familiar classes and overconfidence o…
-
New benchmarks reveal limitations in MLLM visual reasoning capabilities
Researchers are developing new benchmarks and methods to evaluate the visual reasoning capabilities of multimodal large language models (MLLMs). One new benchmark, "The Prestige," uses magic tricks to challenge models' …
-
New framework HUMOR enhances VLM meme generation with hierarchical reasoning
Researchers have developed a novel framework called HUMOR to enhance Vision-Language Models (VLMs) in generating humorous memes. This framework utilizes a hierarchical, multi-path Chain-of-Thought (CoT) approach, allowi…
-
New system optimizes VLM inference on edge SoCs by enabling concurrent accelerator use
Researchers have developed a new system called \"name\" designed to optimize vision-language model (VLM) inference on edge System-on-Chips (SoCs). This system addresses the inefficiency of current VLM runtimes by enabli…
-
New Benchmark Reveals VLM Struggles with Braille Recognition
Researchers have developed BrailleBench, a new benchmark dataset comprising 259 real-world images of Braille, to evaluate the capabilities of Vision-Language Models (VLMs) in recognizing Braille. Their findings indicate…
-
New research reveals VLMs' surprising visual communication and interpretation gaps
A new research paper explores how Vision-Language Models (VLMs) interpret and communicate using visual information. The study found that simple visual symbols, like a red heart, can implicitly influence VLM behavior on …
-
Robot failure diagnosis improved by VLM internal states, study finds
A new research paper explores the challenges of diagnosing failures in bimanual robot tool use, moving beyond single-arm failure analysis. The study introduces a simulated task to evaluate how different sensing channels…
-
New QViD method prunes visual tokens in VLMs for improved efficiency
Researchers have introduced Query-Vision Decomposition (QViD), a new method for pruning visual tokens in large vision-language models (VLMs). This training-free approach leverages the low-rank structure of the query-vis…
-
TrySee pipeline enhances VLM robot task planning in specialized domains
Researchers have developed TrySee, a novel modular pipeline designed to enhance the task planning capabilities of Vision-Language Models (VLMs) in specialized robotic domains. This system integrates general VLM reasonin…
-
New framework uses LLMs to align robot behavior with human intent
Researchers have developed a multimodal framework to align autonomous robots with human intentions using large language models (LLMs) and vision-language models (VLMs). This system translates natural language human purp…
-
RoboReason enhances VLM reasoning for robot manipulation tasks
Researchers have introduced RoboReason, a novel approach to enhance the reasoning capabilities of vision-language models (VLMs) for robotics applications. The study highlights that current state-of-the-art VLMs struggle…