Skip to content

Latest commit

 

History

History
41 lines (36 loc) · 1.59 KB

get_started.md

File metadata and controls

41 lines (36 loc) · 1.59 KB

Data preparation

  1. Prepare the datasets with the download_data.sh script.

    bash data/download_data.sh --path ./data

    You can also download these datasets manually using the URLs in the download_data.sh script. We have provided some alternative download links here, as the download URLs for some files are not functioning.

  2. Download the dataset index files from Google Drive to the split/ folder and then extract them.

    cd split
tar -xf data.tar

    The folder structure for these datasets is shown below.

    VLTVG
├── data
│   ├── Flickr30k
│   │   ├── flickr30k-images
│   ├── other
│   │   ├── images
│   ├── referit
│   │   ├── images
│   │   ├── masks
├── split
│   ├── data
│   │   ├── flickr
│   │   ├── gref
│   │   ├── gref_umd
│   │   ├── referit
│   │   ├── unc
│   │   ├── unc+

Pretrained Checkpoints

We use the pretrained checkpoints provided by TransVG. You can download them here and place them in the pretrained_checkpoints/ folder.