dataset.py provides the general methods to parse the annotations for two datasets,
while other specific scripts, such as vidvrd.py and vidor.py, implements particular details
of the datasets, such as directory structure and label information.
It is recommended to place the downloaded datasets under the same folder as this repository, like:
├── vidor-dataset
│ ├── annotation
│ │ ├── training
│ │ └── validation
│ └── video
├── vidvrd-dataset
│ ├── test
│ ├── train
│ └── videos
└── VidVRD-helper
├── baseline
├── baseline.py
├── dataset
├── evaluate.py
├── evaluation
├── LICENSE
├── README.md
└── visualize.py
Otherwise, you need to change the details in the spefic scripts.
Given a specific dataset, split and task, you can generate a consolidated ground truth file by directly running the specific scripts from the parent directory. For example
python -m dataset.vidor [split] [task] [output_path]