.footer { } Logo Logo
deutsch
/// News
Apple releases Final Cut Pro 10.8 promotional clip

Meta Segment Anything 2 automatically masks objects in videos

[10:52 Thu,1.August 2024   by Thomas Richter]    

Detecting moving objects in videos, automatically masking them, and tracking them across entire video sequences is the foundation for many interesting video effects. Meta&s Segment Anything Algorithm does exactly that—with very high quality. Objects can be selected by the user in real time, and they are then automatically tracked and masked in all subsequent frames of a video. Effects can then be specifically applied to these objects (or the background). The new version can also track objects across multiple settings or even when they are temporarily obscured by another object.

SAM2-titel




A practical use case in video editing is, for example, blurring license plates or faces—SAM2 can pixelate selected objects with minimal effort across multiple scenes. Combined with generative video AIs, there are also many interesting applications for SAM2 for very specific effects. The algorithm can, of course, also be applied to individual images for image editing.


The previous version was also used for scientific purposes, such as in marine science to segment sonar images and analyze coral reefs, in satellite image analysis for disaster relief, and in the medical field for segmenting cell images and assisting in the detection of skin cancer. The new version works in real time and can therefore also be used in live video streams, such as in sports broadcasts.

SAM2-Desaturierung


Fortunately, the improved Segment Anything Model 2 (SAM 2)—like the first version released in April last year—is available under the free Apache 2.0 license for general use and can thus be used free of charge and integrated into other applications (e.g., as a plugin).


SAM 2 surpasses other algorithms for interactive video segmentation in terms of object recognition accuracy and tracking speed. Additionally, it can recognize and mask any object in any video or image (often described as zero-shot generalization), meaning it works even with completely new, previously unseen, and thus untrained objects.

Try it yourself


With the interactive demo, you can test the quality of SAM2 with your own videos in the browser (Safari and Chrome). Objects can be easily selected with a mouse click and then tracked throughout the video. If an object is not recognized correctly or if you only want to mask parts of it, individual areas of a mask can be added or removed in real time. Several video effects are also available, which can be applied to these objects or the rest of the image, such as pixelation or blurring or desaturating the background.

Many users have already done this and posted their results online:

Link more infos at bei ai.meta.com

deutsche Version dieser Seite: Meta Segment Anything 2 - kostenloses Tool maskiert interaktiv Objekte in Videos

  



[nach oben]












Archiv Newsmeldungen

2025

February - January

2024
December - November - October - September - August - July - June - May - April - March - February - January

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000






































deutsche Version dieser Seite: Meta Segment Anything 2 - kostenloses Tool maskiert interaktiv Objekte in Videos



last update : 12.Februar 2025 - 18:02 - slashCAM is a project by channelunit GmbH- mail : slashcam@--antispam:7465--slashcam.de - deutsche Version