Faster image segmentation with Meta’s Segment Anything 2.0

We are thrilled to introduce our enhanced image Auto-Segment tool that takes advantage of Segment Anything 2.0, Meta’s latest foundation model release for image segmentation. The upgrade Auto-Segment tool is available today in the Labelbox image editor. To use, simply choose the AutoSegment icon and start drawing masks at a much faster pace than before!

Segment Anything 2.0 is a state-of-the-art computer vision model that performs better at zero-shot segmentation tasks on both image and video. It builds on the success of Segment Anything by improving speed and accuracy of interactive segmentation tasks.

What is SAM 2?

Segment Anything Model 2 (SAM 2) is a computer vision model developed by Meta. It is designed for real-time promptable object segmentation in both images and videos, achieving state-of-the-art performance. SAM 2 builds on the original Segment Anything Model (SAM) by introducing significant improvements in segmentation accuracy and speed, especially for video content​. This model leverages a new dataset, SA-V, which includes approximately 51,000 real-world videos and over 600,000 masklets, enabling it to segment any object in any visual domain without custom adaptation.

Differences between SAM and SAM 2

The primary differences between SAM and SAM 2 lie in their capabilities and performance enhancements. SAM was groundbreaking for image segmentation, providing accurate results with minimal user interaction. However, SAM 2 extends these capabilities to video segmentation, introducing a unified model that handles both images and videos seamlessly. SAM 2 incorporates a memory mechanism that stores information about the object and user interactions throughout the video, allowing it to maintain accurate segmentation across frames. This results in better video segmentation performance with three times fewer interactions and six times faster processing in images compared to SAM.

Why we should be excited about SAM 2 in Labelbox

The integration of SAM 2 into Labelbox’s image editor is a significant advancement for users who require precise and efficient image segmentation tools. SAM 2’s ability to provide real-time, high-accuracy segmentation for both images and videos means that users can achieve better results with less manual effort. This is particularly beneficial for industries such as autonomous driving, robotics, and content creation, where fast and accurate object segmentation is crucial​. Additionally, the open-source nature of SAM 2 and the extensive SA-V dataset promote broader adoption and innovation in computer vision applications. By incorporating SAM 2 into Labelbox, users can now leverage cutting-edge technology to enhance their workflows and achieve superior segmentation results.

For context, please read the previous article about our AutoSegment functionality: "Auto-Segment 2.0 powered by Meta's Segment Anything Model"

Meanwhile,​​ sign up for a free Labelbox account to try it out, or contact us to learn more.