Baseline benchmark of BOAT-MI

ICCV 2023

This is a subbranch of the Paper [Data-Free Class-Incremental Hand Gesture Recognition], which contains the Pytorch implementation of all the baseline approaches.

Dataset preparation

The data preparation step is the same as what's documented in the repository of our proposed method.

• and replace the dataset directory root_dir in run_trial.sh with your own local dataset directory

for  dataset_name  in ${datasets[*]}; do

if [ $dataset_name  =  "hgr_shrec_2017" ]

then

dataset="hgr_shrec_2017"

root_dir="/ogr_cmu/data/SHREC_2017"

elif [ $dataset_name  =  "ego_gesture" ]

then

dataset="ego_gesture"

root_dir="/ogr_cmu/data/ego_gesture_v4"

fi

Training

Three seeds are randomly picked to run three experiments for each baseline approach.

• You may choose to rerun the whole experiments on your own, but we store the checkpoint of initial pre-trained model for all three seeds to make a fair comparison between different approach. and the saved checkpoint can be download from https://drive.google.com/drive/folders/1gsIPd-BGXvb2zVIWRZwDZ75ejxtxcV05?usp=sharing

  • Place the pre-trained model folder models under the parent directory ogr_cmu
  • Skip this step if you want to run the experiments entirely

• Run all experiments by one command

./scripts/run_experiments_all.sh

• Run single specific experiments by simply changing some configurations in the run_experiments_all.sh file. For example, run ABD approach on Shrec-2017 for one trial.

split_type="agnostic"

CUDA_VISIBLE_DEVICES=0

gpu=0

datasets=("hgr_shrec_2017")

baselines=("Oracle")

trial_ids=(0)

n_trials=${#trial_ids[@]}

n_tasks=1

Acknowledgments

The structure of our code is inspired by Always Be Dreaming.