PyTorch code and the ESSVP dataset for our paper "Adaptive Modality Balanced Online Knowledge Distillation for Brain-Eye-Computer based Dim Object Detection in Aerial Images".
This folder contains the ESSVP multimodal dataset and the experimental code for the AMBOKD-related algorithms To run the experiment:
- Download the essvp_dataset and pretrained uni-modal models from https://www.kaggle.com/datasets/lllll789/ambokd-data, and move them to the ESSVP_dataset and the models folders, respectively.
- Run the following command in the AMBOKD_ESSVP folder to train the model and save the 10-fold cross-validation results:
CMM: python main.py --model_s_fusion CMM --distill None --IS_adjust_lr False --IS_adjust_kd False
MLB: python main.py --model_s_fusion MLB --distill None --IS_adjust_lr False --IS_adjust_kd False
AMM: python main.py --model_s_fusion AMM --distill None --IS_adjust_lr False --IS_adjust_kd False
E-KD: python main.py --model_s_fusion AMM --distill E-KD --IS_adjust_lr False --IS_adjust_kd False
V-KD: python main.py --model_s_fusion AMM --distill V-KD --IS_adjust_lr False --IS_adjust_kd False
MKD: python main.py --model_s_fusion AMM --distill MKD --IS_adjust_lr False --IS_adjust_kd False
EMKD: python main.py --model_s_fusion AMM --distill EMKD --IS_adjust_lr False --IS_adjust_kd False
CA-MKD: python main.py --model_s_fusion AMM --distill CA-MKD --IS_adjust_lr False --IS_adjust_kd False
DML: python main.py --model_s_fusion DML --distill DML --IS_adjust_lr False --IS_adjust_kd False
KDCL: python main.py --model_s_fusion KDCL --distill KDCL --IS_adjust_lr False --IS_adjust_kd False
EML: python main.py --model_s_fusion EML --distill EML --IS_adjust_lr False --IS_adjust_kd False
MMOKD: python main.py --model_s_fusion MMOKD --distill MMOKD --IS_adjust_lr False --IS_adjust_kd False
MMOKD-DG: python main.py --model_s_fusion MMOKD --distill MMOKD --IS_adjust_lr True --IS_adjust_kd False
AMBOKD: python main.py --model_s_fusion MMOKD --distill MMOKD --IS_adjust_lr True --IS_adjust_kd True