Welcome to MMDetection’s documentation!

Get Started

• Prerequisites
• Installation
  • Best Practices
  • Verify the installation
  • Customize Installation
  • Trouble shooting
• Benchmark and Model Zoo
  • Mirror sites
  • Common settings
  • ImageNet Pretrained Models
  • Baselines
  • Speed benchmark
  • Comparison with Detectron2

Quick Run

• 1: Inference and train with existing models and standard datasets
  • Inference with existing models
  • Test existing models on standard datasets
  • Train predefined models on standard datasets
• 2: Train with customized datasets
  • Prepare the customized dataset
  • Prepare a config
  • Train a new model
  • Test and inference
• 3: Train with customized models and standard datasets
  • Prepare the standard dataset
  • Prepare your own customized model
  • Prepare a config
  • Train a new model
  • Test and inference

Tutorials

• Tutorial 1: Learn about Configs
  • Modify config through script arguments
  • Config File Structure
  • Config Name Style
  • Deprecated train_cfg/test_cfg
  • An Example of Mask R-CNN
  • FAQ
• Tutorial 2: Customize Datasets
  • Support new data format
  • Customize datasets by dataset wrappers
  • Modify Dataset Classes
  • COCO Panoptic Dataset
• Tutorial 3: Customize Data Pipelines
  • Design of Data pipelines
  • Extend and use custom pipelines
• Tutorial 4: Customize Models
  • Develop new components
• Tutorial 5: Customize Runtime Settings
  • Customize optimization settings
  • Customize training schedules
  • Customize workflow
  • Customize hooks
• Tutorial 6: Customize Losses
  • Computation pipeline of a loss
  • Set sampling method (step 1)
  • Tweaking loss
  • Weighting loss (step 3)
• Tutorial 7: Finetuning Models
  • Inherit base configs
  • Modify head
  • Modify dataset
  • Modify training schedule
  • Use pre-trained model
• Tutorial 8: Pytorch to ONNX (Experimental)
  • Try the new MMDeploy to deploy your model
  • How to convert models from Pytorch to ONNX
  • How to evaluate the exported models
  • List of supported models exportable to ONNX
  • The Parameters of Non-Maximum Suppression in ONNX Export
  • Reminders
  • FAQs
• Tutorial 9: ONNX to TensorRT (Experimental)
  • Try the new MMDeploy to deploy your model
  • How to convert models from ONNX to TensorRT
  • How to evaluate the exported models
  • List of supported models convertible to TensorRT
  • Reminders
  • FAQs
• Tutorial 10: Weight initialization
  • Description
  • Initialize parameters
  • Usage of init_cfg
• Tutorial 11: How to xxx
  • Use backbone network through MMClassification
  • Use Mosaic augmentation
  • Unfreeze backbone network after freezing the backbone in the config
  • Get the channels of a new backbone
• Tutorial 12: Test Results Submission
  • Panoptic segmentation test results submission
• Tutorial 13: Useful Hooks
  • CheckInvalidLossHook
  • EvalHook and DistEvalHook
  • ExpMomentumEMAHook and LinearMomentumEMAHook
  • NumClassCheckHook
  • MemoryProfilerHook
  • SetEpochInfoHook
  • SyncNormHook
  • SyncRandomSizeHook
  • YOLOXLrUpdaterHook
  • YOLOXModeSwitchHook
  • How to implement a custom hook

Useful Tools and Scripts

• Log Analysis
• Result Analysis
• Visualization
  • Visualize Datasets
  • Visualize Models
  • Visualize Predictions
• Error Analysis
• Model Serving
  • 1. Convert model from MMDetection to TorchServe
  • 2. Build mmdet-serve docker image
  • 3. Run mmdet-serve
  • 4. Test deployment
• Model Complexity
• Model conversion
  • MMDetection model to ONNX (experimental)
  • MMDetection 1.x model to MMDetection 2.x
  • RegNet model to MMDetection
  • Detectron ResNet to Pytorch
  • Prepare a model for publishing
• Dataset Conversion
• Dataset Download
• Benchmark
  • Robust Detection Benchmark
  • FPS Benchmark
• Miscellaneous
  • Evaluating a metric
  • Print the entire config
• Hyper-parameter Optimization
  • YOLO Anchor Optimization
• Confusion Matrix

Notes

• Conventions
  • Loss
  • Empty Proposals
  • Coco Panoptic Dataset
• Compatibility of MMDetection 2.x
  • MMDetection 2.25.0
  • MMDetection 2.21.0
  • MMDetection 2.18.1
  • MMDetection 2.18.0
  • MMDetection 2.14.0
  • MMDetection 2.12.0
  • Compatibility with MMDetection 1.x
  • pycocotools compatibility
• Projects based on MMDetection
  • Projects as an extension
  • Projects of papers
• Changelog
  • v2.26.0 (23/11/2022)
  • v2.25.3 (25/10/2022)
  • v2.25.2 (15/9/2022)
  • v2.25.1 (29/7/2022)
  • v2.25.0 (31/5/2022)
  • v2.24.0 (26/4/2022)
  • v2.23.0 (28/3/2022)
  • v2.22.0 (24/2/2022)
  • v2.21.0 (8/2/2022)
  • Breaking Changes
  • v2.20.0 (27/12/2021)
  • v2.19.1 (14/12/2021)
  • v2.19.0 (29/11/2021)
  • v2.18.1 (15/11/2021)
  • v2.18.0 (27/10/2021)
  • v2.17.0 (28/9/2021)
  • v2.16.0 (30/8/2021)
  • v2.15.1 (11/8/2021)
  • v2.15.0 (02/8/2021)
  • v2.14.0 (29/6/2021)
  • v2.13.0 (01/6/2021)
  • v2.12.0 (01/5/2021)
  • v2.11.0 (01/4/2021)
  • v2.10.0 (01/03/2021)
  • v2.9.0 (01/02/2021)
  • v2.8.0 (04/01/2021)
  • v2.7.0 (30/11/2020)
  • v2.6.0 (1/11/2020)
  • v2.5.0 (5/10/2020)
  • v2.4.0 (5/9/2020)
  • v2.3.0 (5/8/2020)
  • v2.2.0 (1/7/2020)
  • v2.1.0 (8/6/2020)
  • v2.0.0 (6/5/2020)
  • v1.1.0 (24/2/2020)
  • v1.0.0 (30/1/2020)
  • v1.0rc1 (13/12/2019)
  • v1.0rc0 (27/07/2019)
  • v0.6.0 (14/04/2019)
  • v0.6rc0(06/02/2019)
  • v0.5.7 (06/02/2019)
  • v0.5.6 (17/01/2019)
  • v0.5.5 (22/12/2018)
  • v0.5.4 (27/11/2018)
  • v0.5.3 (26/11/2018)
  • v0.5.2 (21/10/2018)
  • v0.5.1 (20/10/2018)
• Frequently Asked Questions
  • Installation
  • Coding
  • PyTorch/CUDA Environment
  • Training
  • Evaluation
  • Model

Switch Language

• English
• 简体中文

API Reference

• mmdet.apis
• mmdet.core
• mmdet.datasets
• mmdet.models
• mmdet.utils

Device Support

• NPU (HUAWEI Ascend)

Indices and tables

• Index

• Search Page