Welcome to MMSegmentation’s documentation!

Get Started

• Prerequisites
• Installation
  • Best Practices
  • Verify the installation
  • Customize Installation
  • Trouble shooting

Dataset Preparation

• Prepare datasets
• Dataset Preparation
• Data Generation

Model Zoo

• Benchmark and Model Zoo
• Model Zoo Statistics

Quick Run

• Train a model
  • Train on a single machine
  • Train with multiple machines
  • Manage jobs with Slurm
• Inference with pretrained models
  • Test a dataset

Tutorials

• Tutorial 1: Learn about Configs
  • Config File Structure
  • Config Name Style
  • An Example of PSPNet
  • FAQ
• Tutorial 2: Customize Datasets
  • Data configuration
  • Customize datasets by reorganizing data
  • Customize datasets by mixing dataset
• Tutorial 3: Customize Data Pipelines
  • Design of Data pipelines
  • Extend and use custom pipelines
• Tutorial 4: Customize Models
  • Customize optimizer
  • Customize optimizer constructor
  • Develop new components
• Tutorial 5: Training Tricks
  • Different Learning Rate(LR) for Backbone and Heads
  • Online Hard Example Mining (OHEM)
  • Class Balanced Loss
  • Multiple Losses
  • Ignore specified label index in loss calculation
• Tutorial 6: Customize Runtime Settings
  • Customize optimization settings
  • Customize training schedules
  • Customize workflow
  • Customize hooks

Useful Tools and Scripts

• Useful tools
  • Get the FLOPs and params (experimental)
  • Publish a model
  • Convert to ONNX (experimental)
  • Evaluate ONNX model
  • Convert to TorchScript (experimental)
  • Convert to TensorRT (experimental)
• Miscellaneous
  • Print the entire config
  • Plot training logs
  • Model conversion
• Model Serving
  • 1. Convert model from MMSegmentation to TorchServe
  • 2. Build mmseg-serve docker image
  • 3. Run mmseg-serve
  • 4. Test deployment
• Confusion Matrix
  • 1.Generate a prediction result in pkl format using test.py
  • 2. Use confusion_matrix.py to generate and plot a confusion matrix
• Model ensemble
  • Usage
  • Description of all arguments
  • Result of model ensemble

Notes

• Changelog
  • V0.30.0 (01/09/2023)
  • V0.29.1 (11/3/2022)
  • V0.29.0 (10/10/2022)
  • V0.28.0 (9/8/2022)
  • V0.27.0 (7/28/2022)
  • V0.26.0 (7/1/2022)
  • V0.25.0 (6/2/2022)
  • V0.24.1 (5/1/2022)
  • V0.24.0 (4/29/2022)
  • V0.23.0 (4/1/2022)
  • V0.22.1 (3/9/2022)
  • V0.22 (3/04/2022)
  • V0.21.1 (2/9/2022)
  • V0.21 (1/29/2022)
  • V0.20.2 (12/15/2021)
  • V0.20.1 (12/14/2021)
  • V0.20 (12/10/2021)
  • V0.19 (11/02/2021)
  • V0.18 (10/07/2021)
  • V0.17 (09/01/2021)
  • V0.16 (08/04/2021)
  • V0.15 (07/04/2021)
  • V0.14 (06/02/2021)
  • V0.13 (05/05/2021)
  • V0.12 (04/03/2021)
  • V0.11 (02/02/2021)
  • V0.10 (01/01/2021)
  • V0.9 (30/11/2020)
  • V0.8 (03/11/2020)
  • V0.7 (07/10/2020)
  • V0.6 (10/09/2020)
  • v0.5.1 (11/08/2020)
• Frequently Asked Questions (FAQ)
  • Installation
  • How to know the number of GPUs needed to train the model
  • What does the auxiliary head mean
  • Why is the log file not created
  • How to output the image for painting the segmentation mask when running the test script
  • How to handle binary segmentation task
  • What does reduce_zero_label work for?

Device Support

• NPU (HUAWEI Ascend)
  • Usage
  • Models Results

Switch Language

• English
• 简体中文

API Reference

• mmseg.apis
• mmseg.core
  • seg
  • evaluation
  • utils
• mmseg.datasets
  • datasets
  • pipelines
• mmseg.models
  • segmentors
  • backbones
  • decode_heads
  • losses

Indices and tables

• Index

• Search Page