README.md

Data

All datasets used by VIGOR live under this single data/ directory. Paths in configs/vigor_talk2car.yaml are resolved relative to the repository root, so place the files exactly as shown below (the actual data is not tracked in git).

Expected layout

Refer to Talk2car for images. Other files are uploaded in Google Drive: data

data/
└── talk2car/
    ├── images/                 # RGB camera frames referenced by ref["file_name"]
    ├── lidar_features/
    │   └── train/              # per-sample LiDAR feature tensors: <name>.pt
    ├── talk2car_key.pt         # dict: sample_token -> LiDAR feature file name
    ├── talk2car_coco.p         # REFER referring expressions (pickle)
    └── talk2car_coco.json      # REFER annotations (COCO-style instances)