open-mmlab/mmengine

Health v2 · maintenance only

Adoption not part of the score

1492 stars · 459 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

MMEngine is the foundational training engine library for OpenMMLab projects, providing a unified training loop, config system, registry, hooks, and logging on top of PyTorch. It powers frameworks like MMDetection, MMSegmentation, and MMPose.

Use cases

• train deep learning models with a unified PyTorch engine
• build custom training loops with hooks and callbacks
• manage experiment configs for computer vision model training
• run distributed training for detection or segmentation models
• standardize training infrastructure across OpenMMLab projects

When to choose

• you are using or extending OpenMMLab frameworks like MMDetection or MMSegmentation
• you want a battle-tested, configurable training engine on PyTorch
• you need hook-based customization of training, validation, and logging

When to avoid

• you want a lightweight plain-PyTorch training script without framework overhead
• your project is not PyTorch-based
• you need a general-purpose ML framework outside the OpenMMLab ecosystem

Facets

library · maturity active

deep-learning machine-learning llm-training deep-learning computer-vision machine-learning python cross-platform pytorch training-loop openmmlab config-system hooks gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="open-mmlab/mmengine")