Accelerated Annotation Supercharged by Meta’s Segment Anything Model

CloudFactory is excited to announce that we've integrated Meta AI’s groundbreaking Segment Anything Model (SAM) into Accelerated Annotation, our best-in-class data labeling and workflow solution, to deliver precise pixel-level segmentation at unprecedented speeds.

When we launched Accelerated Annotation earlier this year, our AI-assisted labeling was already at least five times faster than traditional methods. Now with SAM incorporated into our solution – including its ability to use a variety of input prompts, such as a mouse click – we can provide precise pixel-level segmentation several times faster than before, making it even more accessible to AI and ML teams, and allowing for more attention on label refinement and QA. This will enable more accurate AI models and better performance in a wider range of applications. It is a game changer.

It also means those teams get the benefit of SAM integrated with Accelerated Annotation automated QA workflows, model performance feedback loops, and active learning model to multiply its impact. And, of course, CloudFactory’s world-class workforce with expertise honed over 8 million hours of labeling.

Not just speed, but quality at scale

Our unique combination of a leading AI-assistance tool, workforce management technology, and expert human annotators makes Accelerated Annotation the most accurate and efficient data annotation solution on the market. SAM evolves these capabilities even further.

SAM will replace our existing Level 1 automation tools, which automate the process of extracting the mask for Semantic, Instance, and Panoptic segmentation. Introducing SAM here opens up segmentation for teams that may have deemed it too expensive and time-consuming to work with masks in the past. Due to less label-noise with masks, models should perform much faster than with simpler forms of annotation like bounding boxes. This means that our clients will hit the ground running on new projects with even higher-performing automation.

Through our existing workflows, expert annotators will still be involved at the point of highest impact, like adding classes, tightening edges, and ensuring quality outputs. While SAM is truly excellent, it is still a few steps away from full auto-labeling – for example, it still struggles with high-res images, and there is no classifier involved. This makes CloudFactory’s existing Accelerated Annotation offering, now powered by SAM, the most powerful and complete way to leverage this breakthrough.

Box-to-Instance: Upgrade Your Existing Datasets

Before today, pixel-level segmentation was expensive and time-consuming, leading many AI teams to make do with object detection (bounding boxes) in the past.

Fortunately, you no longer need to be stuck with datasets of hundreds of thousands of images labeled with boxes, where segmentation would provide a significant boost in model performance.

Our Box-to-Instance capability allows us to (very) quickly and easily convert bounding boxes to instance masks using SAM. We can help you make the most of this breakthrough in AI without losing out on years of investment.

Our new SAM-powered tools are automatically available for all of our Accelerated Annotation clients.

Contact us to learn how we can help you with your data labeling project, including upgrading your existing bounding boxes to instance masks.

If you are building your annotation strategy and need more details on CloudFactory's approach to AI-assisted labeling, download our white paper: Accelerated Data Labeling: A Comprehensive Review of Automated Techniques.