README.md

In this work, we present Recursive fusion of Joint Cross-Attention across audio and visual modalities for person verification.

References

If you find this work useful in your research, please consider citing our work 📝 and giving a star 🌟 :

@article{praveen2024audio,
  title={Audio-Visual Person Verification based on Recursive Fusion of Joint Cross-Attention},
  author={Praveen, R Gnana and Alam, Jahangir},
  journal={arXiv preprint arXiv:2403.04654},
  year={2024}
}

There are three major blocks in this repository to reproduce the results of our paper. This code uses Mixed Precision Training (torch.cuda.amp). The dependencies and packages required to reproduce the environment of this repository can be found in the environment.yml file.

Creating the environment

Create an environment using the environment.yml file

conda env create -f environment.yml

Models and Text Files

The pre-trained models of audio and visual backbones are obtained here

The fusion models trained using our fusion approach can be found here

The text files can be found here

train_list :  Train list
val_trials :  Validation trials list
val_list : Validation list
test_trials : VoX1-O trials list
test_list : Vox 1-O list

Table of contents

• Preprocessing
  • Step One: Download the dataset
  • Step Two: Preprocess the visual modality
• Training
  • Training the fusion model
• Inference
  • Generating the results

Preprocessing

Return to Table of Content

Step One: Download the dataset

Return to Table of Content Please download the following.

• The images of Voxceleb1 dataset can be downloaded here

Step Two: Preprocess the visual modality

Return to Table of Content

• The downloaded images are not properly aligned. So the images are aligned using Insightface The preprocessing scripts are provided in preprocessing folder

Training

Return to Table of Content

• sbatch run_train.sh

Inference

Return to Table of Content

• sbatch run_eval.sh

👍 Acknowledgments

Our code is based on AVCleanse