MMdetection is not supported anymore

Whole set of MM* tools is on EOL...

Page content

I’ve used MMDetection (mmengine, mdet, mmcv) quite a bit, And now looks like it’s out of the game. It’s a pity. I liked it’s model zoo.

See, for example here: Concrete Reinforcement Bar Caps object detection with tensorflow and here: Training Object Detector AI with Label Studio & MMDetection

mm train

mmdetection is EOL

I was curious why there was no releases during last year and here is why:

As we see in https://github.com/open-mmlab/mmdetection/discussions/11815

mchaniotakis is commented on Jun 28, 2024:

I think that, even though openmmlab has a steep learning curve,
once its set up is an amazing tool to work with.

However the development seems to have stoped since the end of December
after the passing of Professor Tang Xiaoou.

Are there plans to continue the development in the future
(or atleast assign new maintainers)?

And the repply was from maisonhai3:

They droped the MMLab.
MMLab's head professor passed years ago.
Then, they move staffs to InternLM.
Then, even the InternLM is half-dead now.

I love the MMLab works. Their code quality is greate. Easy to maintain.

A bit about MMDetection…

MMDetection is an open-source object detection toolbox developed by OpenMMLab, based on PyTorch. It provides a comprehensive framework for tasks such as object detection, instance segmentation, and panoptic segmentation. MMDetection is modular, allowing users to customize components such as backbones, necks, heads, and loss functions to build single-stage, two-stage, or multi-stage detection models.

Key features include:

  • Modular Design: Components like Backbone, Neck, DenseHead, ROIExtractor, and ROIHead can be customized or replaced.
  • Rich Model Support: Includes state-of-the-art models like Cascade R-CNN, FCOS, and Dynamic R-CNN.
  • Integration: Compatible with tools like ArcGIS for streamlined workflows.
  • Flexibility: Supports training and inference with custom configurations and pretrained weights.

MMDetection is widely used in computer vision research and applications due to its flexibility and performance benchmarks.