Style-aware two-stage learning framework for video captioning

Dependencies

• Python 2.7
• Pytorch
• Microsoft COCO Caption Evaluation
• CIDEr
• numpy, scikit-image, h5py, requests

You can use anaconda or miniconda to install the dependencies:

conda create -n savc python=2.7 pytorch=1.0 scikit-image h5py requests
conda activate savc

Installation

First clone this repository

git clone --recursive https://github.com/myccver/SAVC.git

Then, please run following script to download Stanford CoreNLP 3.6.0 models into coco-caption/:

cd msrvtt/coco-caption
./get_stanford_models.sh

Stage-1: Generate captions in different styles

Training

MSR-VTT

cd msrvtt
sh train.sh

MSVD

cd msvd
sh train.sh

Testing

MSR-VTT

cd msrvtt
sh test.sh

MSVD

cd msvd
sh test.sh

Stage-2: Style aware from video content

This part of the code is not released.