SAM 3 - AI at Meta

2025-07-21 · Source: ai.meta.com via Google News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Image Processing · Depth: Advanced, short

Summary

Meta has introduced Segment Anything Model 3 (SAM 3), an advanced AI model capable of identifying, segmenting, and tracking any object in images or videos using text and visual prompts. This new iteration builds upon its predecessors, SAM 1 and SAM 2, by integrating open-vocabulary text and exemplar prompts alongside existing click-based visual segmentation. SAM 3 achieves state-of-the-art performance across various segmentation tasks, including document, aerial, flora and fauna, industrial, medical, and sports benchmarks. It is designed for real-world applications, with planned integration into Instagram Edits for video creation and the Meta AI app's Vibes feature. The model is built on a unified, promptable architecture, leveraging a large-scale, diverse training dataset and a powerful perception encoder.

Key takeaway

For Machine Learning Engineers developing computer vision applications, SAM 3 offers enhanced capabilities for object segmentation and tracking. Its multimodal prompting (text, exemplar, visual) and state-of-the-art performance across diverse benchmarks can streamline development workflows and improve accuracy. Consider integrating SAM 3 to enable more intuitive and precise object manipulation in your image and video processing pipelines, especially for interactive editing or content creation tools.

Key insights

SAM 3 offers state-of-the-art object segmentation and tracking across images and videos using diverse text and visual prompts.

Principles

• Unified architecture for multimodal prompting
• Iterative refinement improves segmentation accuracy

Method

SAM 3 employs a unified, promptable model architecture, trained on a large, diverse dataset with a powerful perception encoder, to enable segmentation via language, exemplars, and visual prompts.

In practice

• Use text prompts to mask objects by description
• Draw boxes for exemplar-based object segmentation
• Apply positive/negative clicks for interactive refinement

Topics

• SAM 3
• Object Segmentation
• Video Tracking
• Text Prompts
• Visual Prompts

Code references

• facebookresearch/sam3

Best for: Machine Learning Engineer, Research Scientist, AI Scientist, AI Engineer, Computer Vision Engineer

Related on AIssential

• Meta's SAM 3 (Free + Open Source) — SAM 3 enables text-prompted, open-source video object segmentation for easy analysis.
• SAM 3: Building a unified model architecture for detection and tracking — SAM 3 unifies object detection and tracking using distinct transformers for each task.
• SAM 3: Concept-Based Visual Understanding and Segmentation — SAM 3 unifies detection, segmentation, and tracking of open-vocabulary concepts using text or visual prompts.
• SAM 3 for Video: Concept-Aware Segmentation and Object Tracking — SAM3 unifies detection, segmentation, and tracking in video by maintaining streaming memory and tracking state across frames.
• SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning - AI at Meta — SAM 3.1 enhances real-time video object tracking via multiplexing, doubling processing speed and reducing GPU demands.
• Grounding Qwen3-VL Detection with SAM2 — Combining Qwen3-VL's detection with SAM2's segmentation enables robust, natural language-driven object grounding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ai.meta.com via Google News.