MLCVZoo MMDetection

The MLCVZoo is an SDK for simplifying the usage of various (machine learning driven) computer vision algorithms. The package mlcvzoo_mmdetection is the wrapper module for the mmdetection framework.

Install

NOTE: For some panoptic segmentation models of mmdetection you need to have the panopticapi installed. How this is handled in mlcvzoo-mmdetection is listed below. Otherwise you will get an ImportError.

Install for developer

Clone repository

cd YOUR_PATH
git clone git@git.openlogisticsfoundation.org:silicon-economy/base/ml-toolbox/mlcvzoo-models/mlcvzoo-mmdetection.git
cd mlcvzoo-mmdetection

Create and install python environment

virtualenv .venv
source .venv/bin/activate

# Install dependencies with all extras (including panopticapi)
source build.sh

For VSCode IDE add the relevant environment variables in .venv/bin/activate.

export PYTHONPATH=PARENT_OF_YOUR_REPOSITORY
# The MMDETECTION_DIR is needed in our unit tests for defining the
# path to the configuration files of mmdetection. Since mmdet v2.15.0
# the configuration are stored in the mmdet/.mim/ folder of the installed
# mmdet package in your python environment. For older versions a specific checkout
# is needed.
export MMDETECTION_DIR="$VIRTUAL_ENV/lib/python3.10/site-packages/mmdet/.mim/"

If you want to use the complete functionality of this module including model deployment, you must additionally clone the mmdeploy framework. After cloning the repository, add the environment variable MMDEPLOY_DIR to .venv/bin/activate and set it to the path of the mmdeploy repository.

Install for package users

The following command installs the mlcvzoo-mmdetection package, however we recommend to have a look at the build.sh script. It is gathering all fixes that are needed to setup a running python environment.

pip install mlcvzoo-mmdetection

NOTE: In order to use panoptic segmentations you have to install the following, as specified by mmdetection:

pip install panopticapi @ git+https://github.com/cocodataset/panopticapi.git@7bb4655548f98f3fedc07bf37e9040a992b054b0

Further documentation

• Ar42 Documentation
• Installation guide
• Configuration Usage

Technology stack

• Python