cmds.txt

#########  Stage 1  ##########
Download SBD dataset from http://home.bharathh.info/pubs/codes/SBD/download.html; then extract the dataset directory to current directory.

python S0_load_mat.py

which will convert the original mat file to the desired format for remaining stages.
#########  Stage 1  ##########

# degrade to original method by setting max_depth = 0 and mode = ese_ori
python main.py --stage encoding --mode hierarchy_encoding --datalist train --max_depth 0 --process_mode ese_ori --save

# for coco
python main.py --stage encoding --mode hierarchy_encoding --dataset coco --datalist train --max_depth 0 --process_mode ese_ori --save

# Our method
python main.py --stage encoding --mode hierarchy_encoding --datalist train --max_depth 5 --process_mode hybrid --save


#########  Stage 2  ##########

# original methd + svd
python main.py --stage factorization --mode hierarchy_encoding --datalist train --max_depth 0 --process_mode ese_ori --save

# Our method + svd
python main.py --stage factorization --mode hierarchy_encoding --datalist train --max_depth 5 --process_mode hybrid --save

# Our method + fms
python main.py --stage factorization --mode hierarchy_encoding --datalist train --max_depth 5 --process_mode hybrid --fms --dim 50 --save


#########  Stage 3  ##########

# original methd
python main.py --stage convert --mode hierarchy_encoding --datalist train --max_depth 0 --process_mode ese_ori --save --dim 8

# Our method
python main.py --stage convert --mode hierarchy_encoding --datalist train --max_depth 5 --process_mode hybrid --save --dim 8

# Our method with fms
python main.py --stage convert --mode hierarchy_encoding --datalist train --max_depth 5 --process_mode hybrid --save --method fms --dim 8