MMDetection Base

Simplified repository for building MMDetection-based repository

Files

In advance, the basic tool files have been added from MMDetection.

├── LICENSE
├── README.md
├── configs
├── docker
│   ├── Dockerfile
│   ├── requirements_core.txt
│   └── requirements_dev.txt
├── docker-compose.yml
├── {src}  # custom module
├── tests  # test for custom module
└── tools
    ├── test.py
    └── train.py

Installation

In this repository, MMDetection is installed by docker. MMDetection files (excluding tools) are not included to build a repository that keeps up with the latest version of MMDetection.

Please specify some versions in requirements_core.txt and docker-compose.yml.

Current versions:

• python 3.10.6
• pytorch 1.12.0
• torchvision 0.13.0
• mmcv 1.7.0
• mmdetection 2.25.3

Build & run container:

export UID=$(id -u)
docker compose build dev
docker compose run --rm -v {data root}:/data dev

Usage

configs

Please add the mmdet config files to use from MMDetection.

run

train:

python tools/train.py {config file}

test:

python tools/test.py {config file} {ckpt file}

case 1. Training with custom dataset

The simple usage for custom dataset (faster rcnn, yolo):

docs/simple_training_with_custom_dataset.md

License

This repository itself for building repositories is MIT licensed, but the built repositories are subject to the MMDetection license.