Back to Gap Map Infrastructure / Classic ML & computer vision

MMDetection

OpenMMLab
open source / Overall score: 3.4

MMDetection is OpenMMLab's PyTorch toolbox for object detection, instance segmentation and panoptic segmentation, built from interchangeable modules on the MMEngine and MMCV libraries. Its model zoo covers two-stage, one-stage and transformer detectors such as Faster R-CNN, RTMDet and DINO, along with open-vocabulary detectors such as GLIP and MM-Grounding-DINO, an open reproduction of Grounding DINO that finds objects named in a text prompt.

Openness

5 high confidence
5.0
license
Apache-2.0 (OSI; code)
source
public (github.com/open-mmlab/mmdetection)
core features withheld
no — OpenMMLab research toolbox; nothing sold

The code is Apache-2.0 and builds from the public repository, and OpenMMLab sells nothing around it. The last tagged release was v3.3.0 in January 2024, and the repository has not been pushed to since August 2024.

Adoption

3 high confidence
3.0

PyPI downloads of the mmdet package over the trailing 30 days. A download count includes CI jobs and mirrors as well as people, so it measures installs rather than users.

Capability

4 medium confidence
4.0

MMDetection runs detectors that take the object names as text at inference, so a new label set needs no training run, as Ultralytics does with YOLO-World and YOLOE. Most of its model zoo is still trained on fixed benchmark classes.

Verified 2026-11-12