Visually-Aware Audio Captioning Transformer (V-ACT)

This repository contains source code for our paper Visually-Aware Audio Captioning with Adapeive Audio-Visual Attention.

Set up environment

Create a conda environment with dependencies: conda env create -f environment.yml -n name
All of our experiments are running on RTX 30 series GPUs with CUDA11. This environment may just work for RTX 30x GPUs.

Set up dataset

All the experiments were carried out on AudioCaps dataset, which is sourced from AudioSet. Our download version （video available) contains 47745/49837 audio clips in training set, 480/495 audio clips in validation set, 928/975 audio clips in test set.

For reproducibility, our downloaded version can be accessed at:

Baidu password: a1p4
Google Drive

To prepare the dataset:

Put downloaded zip files under data directory, and run data_unzip.sh to extract the zip files.
Run python data_prep.py to create h5py files of the dataset.

Prepare evaluation tool

Run coco_caption/get_stanford_models.sh to download the libraries necessary for evaluating the metrics.

Experiments

Training

The default setting is for 'ACT_m_scratch'
Run experiments: python train.py -m audio -n exp_name
Set the parameters you want in settings/settings.yaml
Modality [only audio available now]:
- audio
- video
- audio-visual

Pretrained encoder

We provide two pretrained encoders, one is a pretrained DeiT model, another is the DeiT model pretrained on AudioSet.

To use pretrained encoder:

Download the pretrained encoder model and put it under the directory pretrained_models
Set settings in settings/settings,yaml
- set encoder.model: to 'audioset'
- set encoder.pretrained to 'Yes'
- set path.encoder to the model path, e.g. 'pretrained_models/audioset_deit.pth'
Run experiments

Cite

If you wish to cite this work, please kindly cite the following paper:

@article{liu2022visually,
  title={Visually-Aware Audio Captioning With Adaptive Audio-Visual Attention},
  author={Liu, Xubo and Huang, Qiushi and Mei, Xinhao and Liu, Haohe and Kong, Qiuqiang and Sun, Jianyuan and Li, Shengchen and Ko, Tom and Zhang, Yu and Tang, Lilian H and others},
  journal={arXiv preprint arXiv:2210.16428},
  year={2022}
}

liuxubo717/V-ACT