YOLOR

This fork is just a packaged version of github.com/WongKinYiu/yolor (which is the implementation of the paper You Only Learn One Representation: Unified Network for Multiple Tasks)

Install using pip install git+https://github.com/marcinwrochna/yolor.git

Usage:

from yolor_mimsolutions.models.models import Darknet
from yolor_mimsolutions.utils.datasets import letterbox
...

The scripts can be run from the module as well, for example:

python -m yolor_mimsolutions train --batch-size 32 \
                                   --img 960 960 \
                                   --name ${RUN_NAME} \
                                   --data ${DATASET}/dataset_config.yaml \
                                   --cfg yolor_p6.cfg \
                                   --hyp hyp.scratch.1280.yaml \
                                   --weights '' \
                                   --device 7 \
                                   --epochs 300

Files like yolor_p6.cfg or hyp.scratch.1280.yaml are then retrieved from inside the package (or you can provide your own).

Modifying

After modifying the repo, you might need to rebuild it (I'm actually not sure):

rm -rf yolor_mimsolutions.egg-info/ dist/
python -m build 
git add dist