DATA.md

March 3, 2023 ยท View on GitHub

๐Ÿ€ DATA Preparation

Pre-training

  1. Download WebVid-2M
  2. Gererate a json file from WebVid-2M annotation. We provide the pretraining annotation here
    {
        YOUR VIDEO PATH: CAPTION
        YOUR VIDEO PATH: CAPTION
        YOUR VIDEO PATH: CAPTION
        ....
        YOUR VIDEO PATH: CAPTION
        YOUR VIDEO PATH: CAPTION
    }

Fine-tuning

  1. Download MSRVTT
mkdir data
cd data
wget https://www.robots.ox.ac.uk/~maxbain/frozen-in-time/data/MSRVTT.zip  
unzip MSRVTT.zip 
  1. Split the train/test set following the Frozen in time. Gererate the train/test json file from MSRVTT. We provide our splited annotation here
    {
        YOUR VIDEO PATH: CAPTION
        YOUR VIDEO PATH: CAPTION
        YOUR VIDEO PATH: CAPTION
        ....
        YOUR VIDEO PATH: CAPTION
        YOUR VIDEO PATH: CAPTION
    }