First of all, clone the custom yolact++ repository, which contains the setup.py file, in a chosen directory (this should not change after installation). This repository has been slightly modified by the Robolab Leonardo fellows. Python 3 is needed to run Yolact++.
To install Yolact++ on your pip environment, activate your environment (or install directly on system if you prefer) and run:
cd YOUR_YOLACT_PATH/yolact
pip install -e .
Moreover, to use Yolact++ you need to install DCNv2. There exists two versions of DCNv2 in the Yolact++ repository (DCNv2 and DCNv2_latest directories). Choose the version you need to use:
- DCNv2 is used for OLDER GPU architectures (compatible with older pytorch version)
- DCNv2_latest is used for NEWER GPU architectures (compatible with latest pytorch version)
To install it, substitute YOUR_DCNv2_FOLDER with DCNv2 or DCNv2_latest in the following:
cd YOUR_YOLACT_PATH/yolact/external/YOUR_DCNv2_FOLDER
pip install -e .