MMCV is the open-source computer vision foundation library from OpenMMLab, providing the shared building blocks used across the OpenMMLab ecosystem. It has collected about 6,500 stars on GitHub, is licensed under Apache-2.0, and remains actively maintained with updates in 2026.
MMCV sits underneath well-known OpenMMLab projects such as MMDetection and MMSegmentation, supplying the common machinery they all need: configuration and registry systems, data loading and transformation pipelines, training hooks, logging, and optimized operators for vision workloads. Instead of every vision project reinventing these pieces, MMCV standardizes them, which makes experiments easier to reproduce and code easier to share. It supports both PyTorch-based training workflows and deployment-oriented utilities, and it is developed by the OpenMMLab community with extensive documentation.
MMCV’s design emphasizes modularity and reproducibility. Its registry system lets researchers swap backbones, losses and data pipelines through configuration files instead of code changes, which is a big reason the OpenMMLab ecosystem produces so many comparable, reproducible results in papers. The library runs on Linux, Windows and macOS, supports CPU and GPU training, and its operator library includes components for tasks such as deformable convolution, rotated object detection and point-cloud processing that are painful to implement from scratch.
MMCV is free and open source under the Apache-2.0 license. There is no commercial edition or paid support tier required; the full library is available to everyone through GitHub and PyPI.
MMCV is for computer vision researchers and engineers building on the OpenMMLab stack, or anyone who wants a battle-tested foundation for training and deploying vision models in PyTorch.