NEW! Label Studio 1.9.0Introducing Autocomplete for Labeling Configurations!

Integration Spotlight: OpenMMLab & Label Studio

Erin Mikail Staples

Senior Developer Community Advocate

OpenMMLab (short for Open Media and Machine Learning Laboratory) is an open source organization advancing the field of computer vision and multimedia understanding. They are mainly focused on research and tools for computer vision, enabling developers and data scientists of all types to build tools that rely on object detection, image segmentation, pose estimation, and video understanding.

As a leader in the open source machine learning community, we’re thrilled to have an active and openly-maintained integration with OpenMMLab.

How the integration works

Use OpenMMLab with Label Studio through our machine learning API to provide semi-automated bounded-box labeling using the mmdetection library and the RTMDdet model.

(Want some help getting started with Label Studio? Check out this tutorial.)

What this looks like in practice

Data scientists can run pre-labeling on images by integrating the standard COCO format with MMDet’s existing algorithms and weight files.

Once the model is loaded into Label Studio, you can begin running predictions and enable auto-annotation speeding up your process.

Once auto-annotation is enabled, you’ll be given the option to label images one-by-one with the model predicting the bounding box for each image.

These tools combine to create efficient image-labeling workflows within the same ecosystem. Thanks to Label Studio’s model refinement feature, users can create a high-quality dataset with minimal effort compared to the extensive work that can come with manually annotating images.

Why we ❤ it!

OpenMMLab and Label Studio combine to create a powerful combination saving you time and energy across not only the labeling process but also improvements and further model refinement.

Here are just some of the perks that come with integrating Label Studio with OpenMMLab:

  • Automated Labeling: get a quick start on your image labeling with automatic object detection
  • Standard COCO Format: bounding boxes are supplied using the standard COCO format
  • Fast Integration: The library provides the machine learning connection, allowing for a rapid connection to Label Studio with minimal coding
  • Open Sourced: OpenMMLab is a very active open source community focused on continuously improving and extending research within computer vision.

Community Contributions

The OpenMMLab team are active contributors and participants in the Label Studio community Slack, and community contributors have extended the use of OpenMMLab by building on top of it as well.

For example, check out this playground feature showcasing Segment Anything in combination with OpenMMLab with both Point2Label and Bbox2Label. This added integration allows annotators to simply click a point within a specific object area to get the mask and bounding box annotation.

Shoutout to Label Studio community member Shivansh S. for the work put into the Segment Anything integration.

You also might like…

Are you a fan of the OpenMMLab integration? Are you working to develop other computer vision toolsets? You also might like some of our other integrations, including TensorFlow, YOLO Image Format, or PyTorch.

If you’re an engineer, data scientist, or just curious to learn more about computer vision or object detection, you may also want to check out SciKit Image, Hugging Face’s State of Computer Vision, this paper on training diffusion models with RLHF, or the book Deep Learning: A Visual Approach from Starch Press.

Related Content