data_preparation.md

Data Preparation

We provide some tips for MMAction2 data preparation in this file.

• Data Directory Structures
• Getting Data
  • ActivityNet Captions
  • TACoS
  • MAD

Data Directory Structures

./data
    /<dataset_name_1>
        /feature
            *.hdf5
            ...
        train.json
        val.json
        test.json
    ...

Getting Data

Please follow the guide below to prepare the datasets.

ActivityNet Captions

• Please download the annotations and pre-extracted features.
• Put all files into ./data of the repository.

TACoS

• Please download the annotations and pre-extracted features.
• Put all files into ./data of the repository.

MAD

• Follow the instructions in MAD repository to download the dataset.
• Using the script tools/data/mad/convertion.py to convert the data into our standard format.