README.md

Experiments with the OS2D methods (retail and INSTRE datasets)

Preparations

# activate the env
conda activate os2d
# move to the root folder
# set OS2D_ROOT, e.g., by OS2D_ROOT=`pwd`
cd $OS2D_ROOT
export PYTHONPATH=$OS2D_ROOT:$PYTHONPATH

Train models

# to use one local GPU run
python experiments/launcher_exp1.py
python experiments/launcher_exp2.py
python experiments/launcher_exp3_instre.py
# note that the first call will process the INSTRE dataset and create the cache file, this might cause crashes if done by deveral proceses in parallel, use --job-indices flag to run only some jobs first

View logged information

# View all the saved logs in Visdom
python os2d/utils/plot_visdom.py --log_paths output/exp1
python os2d/utils/plot_visdom.py --log_paths output/exp2
python os2d/utils/plot_visdom.py --log_paths output/exp3

Collect data for ablation tables

# Table 1:
python experiments/launcher_exp1_collect.py
# Table 2:
python experiments/launcher_exp2_collect.py

Evaluation on test sets

# Retail product datasets: run eval
python experiments/launcher_grozi_eval.py
# Collect results (create a part of Table 3)
python launcher_grozi_eval_collect.py

# INSTRE datasets: run eval
python experiments/launcher_instre_eval.py
# Collect results (create a part of Table 4)
python launcher_instre_eval_collect.py