Segment Anything Model 3
PulseAugur coverage of Segment Anything Model 3 — every cluster mentioning Segment Anything Model 3 across labs, papers, and developer communities, ranked by signal.
- 2026-05-22 product_launch Meta AI launched Segment Anything Model 3 (SAM 3), an advanced model for object detection, segmentation, and tracking. source
7 day(s) with sentiment data
-
Young team wins embodied AI challenge by collecting own data in 48 hours · 1 source tracked
A team named TARS, with an average age of 25, won the first Embodied Large Model Challenge hosted by Ant Group's Lingbo. The team, led by a postgraduate student, impressed judges by collecting their own dataset from scr…
-
New PVIS-10K dataset enhances drone-based rooftop solar panel inspection
Researchers have introduced PVIS-10K, a new dataset designed to improve rooftop photovoltaic inspection using drone imagery. The dataset comprises over 10,000 high-resolution images with detailed annotations for more th…
-
New framework maps solar panels with high fidelity using AI and UAVs
Researchers have developed a novel training-free framework called MV-PVMapper for high-fidelity photovoltaic module mapping using multi-view UAV imagery. This system leverages the Segment Anything Model 3 (SAM 3) for in…
-
AgentCount workflow automates microbial colony counting using LLMs and SAM 3
Researchers have developed AgentCount, an open-source agentic workflow designed to automate colony counting from agar plate images, a process crucial for food safety and microbiological studies. This workflow utilizes t…
-
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…
-
New CASE benchmark evaluates AI segmentation models with positive and negative visual prompts
Researchers have introduced Specified Visual Prompt Segmentation (SVPS), a new framework for few-shot semantic segmentation that utilizes both positive and negative visual exemplars to define target concepts. To evaluat…
-
Text-steerable visual encoders exhibit social bias, researchers find
Researchers have identified significant social biases in text-steerable visual encoders, specifically within the SteerViT model. These models, designed to point to image regions based on text prompts, were found to exhi…
-
Unsloth releases v0.1.911-beta with scheduled tasks and FLUX 3 support
Unsloth has released version 0.1.911-beta, introducing scheduled tasks for running prompts locally or via API. This update also brings support for FLUX 3 image editing, performance enhancements for Mac users, and the in…
-
New visual encoder SteerViT shows racial and gender bias, researchers find
Researchers have identified significant social biases in the SteerViT model, a text-steerable visual encoder. When prompted with demographic or crime-related terms, SteerViT disproportionately associates these with spec…
-
AI pipeline automates sugar beet detection using SAM 3 on edge devices
Researchers have developed a new pipeline for detecting sugar beets, even when partially occluded, using an image-based approach. This system leverages the Segment Anything Model 3 (SAM 3) for automatic annotation, redu…
-
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…
-
SAM3RL uses reinforcement learning to boost visual object tracking
Researchers have developed SAM3RL, a new extension for the Segment Anything Model 3 (SAM 3) that enhances visual object tracking. This model utilizes reinforcement learning to control its memory bank, deciding which fra…
-
New Anaximander system simplifies geospatial deep learning model integration
Researchers have developed Anaximander, an open-source system designed to simplify the use of deep learning models for geospatial analysis. The system addresses the challenge of integrating diverse models and compute ba…
-
New dataset benchmarks AI for wild bird detection
Researchers have developed a new benchmark dataset for detecting Western Bluebirds in the wild, addressing challenges like small bird size, background clutter, and variable lighting. The dataset includes over 6,000 labe…
-
New FaVOS benchmark highlights flaws in video object segmentation evaluation
A new benchmark called FaVOS has been introduced to evaluate Video Object Segmentation (VOS) methods, particularly in scenarios where objects appear intermittently. Researchers found that existing evaluation metrics lik…
-
Dyna3 framework enables training-free 4D dynamic scene reconstruction
Researchers have developed Dyna3, a novel framework that enables dynamic 4D scene reconstruction using depth foundation models without requiring any fine-tuning. This method leverages the implicit motion-discriminative …
-
HyperSAM: A New Foundation Model for Hyperspectral Remote Sensing
Researchers have developed HyperSAM, a novel foundation model designed for hyperspectral remote sensing. This model addresses limitations in existing datasets by synthesizing high-resolution hyperspectral cubes from mul…
-
New methods boost open-vocabulary semantic segmentation for specialized domains
Researchers have developed two novel approaches to enhance open-vocabulary semantic segmentation (OVSS) for specialized domains. One method, Preference-Guided Adaptation, utilizes prompt disagreement to generate prefere…
-
Feyn releases MultiMatte, a promptable image background removal model
Feyn has released MultiMatte, a new image background removal model that uses natural language prompts to identify and isolate objects. Built upon Meta's Segment Anything Model 3 (SAM 3), MultiMatte employs low-rank fine…
-
LeCor method improves 3D lung tumor segmentation with meta-learned training
Researchers have developed LeCor, a novel method for improving 3D lung tumor segmentation in computed tomography (CT) scans. LeCor utilizes meta-learned test-time training, where each clinician's correction acts as a tr…