The medvidcl from lireanstar

medvidcl's Introduction

A Two-Stage Cross-modal Fusion Method for

Medical Instructional Video Classification

一种用于医学教学视频分类的两阶段交叉模式融合方法

First Place in BioNLP-22 Medical Video Classification

Method

Code

Please download the I3D features from Baidu disk first, and then unzip *.npy file and put it in /code/data/features/I3D.

Deberta-v3-large model is used by default, or birdbird model can be replaced

The model and experimental records after training will be saved in the /code/paperlog folder.

One-Stage with language

cd code
python Train_One.py

Two-Stage with language

cd code
python Train_Two_1.py
python Train_Two_2.py

One-Stage with MultiModal

cd code
python Train_One_TV.py

Two-Stage with MultiModal

cd code
python Train_Two_1_TV.py
python Train_Two_2_TV.py

Result on offline

Online Result

Our scheme achieved SOTA performance.

Citation

Please feel free to cite our [paper]{https://aclanthology.org/2022.bionlp-1.21/).

@inproceedings{li2022vpai_lab,
  title={Vpai\_lab at medvidqa 2022: A two-stage cross-modal fusion method for medical instructional video classification},
  author={Li, Bin and Weng, Yixuan and Xia, Fei and Sun, Bin and Li, Shutao},
  booktitle={Proceedings of the 21st Workshop on Biomedical Language Processing},
  pages={212--219},
  year={2022}
}

Recommend Projects