RefCOCO+
PulseAugur coverage of RefCOCO+ — every cluster mentioning RefCOCO+ across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
New quantization method improves Segment Anything Model 3 efficiency
Researchers have developed a new post-training quantization technique called Semantics-Preserving Quantization (SPQ) to reduce the memory and computation costs of the Segment Anything Model 3 (SAM 3). This method aims t…
-
LOCUS model improves visual grounding with draft-and-verify approach
Researchers have developed a new visual grounding model called LOCUS that improves accuracy by using a draft-and-verify approach. The model first generates a speculative bounding box using a regression head and then ref…
-
LaMER framework enhances weak supervision for image region localization
Researchers have introduced LaMER, a novel framework for weakly supervised Referring Expression Comprehension (REC). LaMER addresses the challenge of localizing image regions based on natural language descriptions by dy…
-
LLMSeg pipeline uses LLMs for image segmentation without training
A new training-free pipeline called LLMSeg has been developed that leverages multimodal large language models (MLLMs) for image segmentation. This approach bypasses the need for dedicated segmentation architectures or t…
-
Xiaomi-Lens-0 model enhances image region understanding using Qwen 3.5
Researchers have introduced Xiaomi-Lens-0, a 4B multimodal model designed to improve fine-grained region understanding in images. Built upon the Qwen 3.5 architecture, Xiaomi-Lens-0 predicts masks from its internal visu…
-
DVD method enhances MLLM perception with efficient vector decoding
Researchers have introduced Dynamic Vector Decoding (DVD), a novel method designed to enhance the efficiency of multimodal large language models (MLLMs) in perception tasks. DVD addresses limitations in current approach…
-
New research tackles referring expression segmentation with novel policies and benchmarks
Two new research papers published on arXiv introduce novel approaches to referring expression segmentation (RES), a task that involves linking natural language descriptions to pixel-level object masks. The first paper, …
-
LAS-CLIP adapts CLIP visual encoder without parameter changes
Researchers have introduced LAS-CLIP, a novel method for adapting CLIP's visual encoder without altering its original parameters. This approach utilizes a compact MaskAdapter to generate attention biases, which are then…
-
SeGDeP enhances reasoning segmentation by decoupling semantic and geometric prompts
Researchers have developed SeGDeP, a novel interface for reasoning segmentation that disentangles semantic understanding from spatial localization. This approach uses separate branches for semantic prompts and geometric…
-
New WeakMCN Network Improves Referring Expression Tasks
Researchers have developed WeakMCN, a novel multi-task collaborative network designed to improve weakly supervised referring expression comprehension and segmentation. This dual-branch architecture jointly learns both t…
-
New DRAgent framework uses MLLMs for precise object segmentation
Researchers have developed DRAgent, a new framework for Referring Expression Segmentation (RES) that utilizes multimodal large language models (MLLMs). Unlike previous methods that directly predict coordinates, DRAgent …
-
New method boosts vision-language model boundary accuracy without labels
Researchers have developed a novel method called Label-Free Precision Refinement (LFPR) to improve the accuracy of vision-language models in identifying objects and their precise boundaries. This technique allows frozen…
-
New adversarial attack targets SAM3 image segmentation models
Researchers have developed Universal Concept Disruption (UCD), a novel adversarial attack specifically designed to target SAM3 image segmentation models. UCD learns a single image perturbation that can disrupt the model…
-
New Hi-Token method enhances visual grounding accuracy in AI models
Researchers have developed Hi-Token, a novel method for generative visual grounding that improves the accuracy of bounding-box predictions by tokenizing coordinates hierarchically. This approach encodes digits for hundr…
-
StepX-Edge: On-Device UI Vision-Language Model Achieves High Accuracy
Researchers have developed StepX-Edge, a 0.9 billion parameter vision-language model designed for on-device UI understanding. This model addresses the trade-off between accuracy and efficiency on mobile devices through …
-
New SVCR Framework Enhances Weakly Supervised Referring Expression Comprehension
Researchers have developed a new framework called Structured Visual Compositional Representation (SVCR) to improve referring expression comprehension (REC) in weakly supervised settings. This framework explicitly models…
-
HKVLM model improves visual reasoning by separating localization from language
Researchers have developed HKVLM, a novel approach to visual reasoning that separates localization from language generation. This model utilizes a frozen language-aligned detector and a frozen language model, connected …
-
CoLA framework enhances multimodal AI adaptation with dual-path LoRA
Researchers have introduced CoLA (Cross-Modal Low-rank Adaptation), a novel framework designed to efficiently adapt foundation models for multimodal tasks. Unlike existing methods that adapt each modality in isolation, …
-
Researchers seek arXiv endorsement for Locate-SAM2 computer vision paper
Two independent researchers are seeking an endorsement for their paper on a new computer vision system called Locate-SAM2. This system connects NVIDIA's LocateAnything-3B with Meta's SAM 2.1 through a lightweight adapte…
-
New Framework Enhances Semi-Supervised Segmentation with LLM Priors
Researchers have introduced "Learning to Label" (L2L), a novel framework designed to improve semi-supervised referring expression segmentation (SS-RES) by treating pseudo-label generation as a learnable process. L2L uti…