Stacking of strong and weak learners

Data split

Firstly link data to ../data. Then run the following commands.

python make_ensemble_dataset.py --datatrack phase1-main
python make_ensemble_dataset.py --datatrack phase1-ood
python make_ensemble_dataset_wotest.py --datatrack external
python make_ensemble_testphase.py --datatrack phase1-main
python make_ensemble_testphase.py --datatrack phase1-odd

Feature extraction with SSL model

Place the ckpt file of the pretrained model to ../pretrained_model.
Then run the following command.

python extract_ssl_feature.py

Converting results of strong learners for stacking

Place the respective result files to ../strong_learner_result/main1 and ../strong_learner_result/ood1. Then run the following commands.

python convert_strong_learner_result.py phase1-main main1
python convert_strong_learner_result.py phase1-ood ood1
python convert_strong_learner_testphase_result.py testphase-main main1
python convert_strong_learner_testphase_result.py testphase-ood ood1

Stage1

For both main and OOD tracks, run the following command to perform stage1.

./run_stage1.sh

Stage2 and 3 for Main track

Run the following commands.

./run_stage2-3_main.sh # Run stage 2 and 3
./pred_testphase_stage1_main.sh # Predict stage 1
./pred_testphase_stage2-3_main.sh # Predict stage 2 and 3

Stage 2 and 3 for OOD track

Run the following commands.

./pred_stage1_ood.sh # Predict by cross-domain
./run_stage2-3_ood.sh # Run stage 2 and 3
./pred_testphase_stage1_ood.sh # Predict stage 1
./pred_testphase_stage2-3_ood.sh # Predict stage 2 and 3