Skip to content

YOUR_DB.md

Latest commit

 

History

History
44 lines (27 loc) · 1.44 KB

YOUR_DB.md

File metadata and controls

44 lines (27 loc) · 1.44 KB

Custom Dataset

To support your own dataset in our codebase, complete the following steps:

1. Create Dataset Class

Create your dataset class with data loading functionalities and place it in src/data/. This module should load your data on demand. By default, our datasets return two outputs: a torch tensor with video frames, and a dictionary with caption information, such as the raw text, text tokens or other metadata. We recommend sticking to this convention.

2. Support Dataset Loading

In the file src/data/load_data.py, modify the load_data and unwrap_batch_data functions to support loading your dataset.

3. Add your Dataset Config

Add a new configuration file for your dataset in src/configs/datasets/. This is a dictionary that features all the arguments required by your data, as well as the default value for each argument.

4. Test that your Dataset is Recognized

Completing the previous steps should support your dataset. You can test it by running: src/01_create_experiment.py --help. If your dataset is correctly added, it should appear in the list of available datasets.

...
  --dataset_name DATASET_NAME
            Dataset name to add to the exp_params: ['CATER_Easy', 'CATER_Hard', 'CLIPort', 'YOUR_DB']
...

You can now train your our decomposition models and predictors on your own custom dataset.

In case of questions, do not hesitate to open an issue or contact the authors at villar@ais.uni-bonn.de