Prerequisites

Python

pip install -r requirements.txt
chmod 744 ./tools/*.py
python setup.py develop --user

Data

Prepare imagenet data as

/path/imagenet
|_ train
|  |_ n01440764
|  |_ ...
|  |_ n15075141
|_ val
|  |_ n01440764
|  |_ ...
|  |_ n15075141
|_ ...

mkdir -p /path/pycls/pycls/datasets/data
ln -sv /path/imagenet /path/pycls/pycls/datasets/data/imagenet

Command

Modify "results" to specify our dir.

./tools/run_net.py --mode train \
    --cfg configs/energy_effnet/regnetx/RegNetX-32GF_8_gpu.yaml \
    OUT_DIR results \
    LOG_DEST file

If using different number of gpus (other than 8), TRAIN.BATCH_SIZE and OPTIM.BASE_LR need to be modified. Compare RegNetX-32GF.yaml and RegNetX-32GF_8_gpu.yaml.