google-deepmind / google-deepmind/robovqa

Mapping dataset entries to long- and medium-horizon instructions

Open
#1 5 comments 2 reactions 0 assignees View on GitHub
Dominant language
Jupyter Notebook
Stars
37
Forks
4
PR merge metrics
No merged PRs in 30d

Description

Thanks for sharing this great dataset.

I was wondering how it is possible to map each entry of the dataset (as returned by the tf.data.TFRecordDataset or as found in the [json files](https://console.cloud.google.com/storage/browser/gdm-robovqa/json)) back to the medium- and long-horizon [instruction](https://console.cloud.google.com/storage/browser/gdm-robovqa/instructions) it belongs to. For all question types except discriminative negative, the medium-horizon task is contained in the sample itself, but the long-horizon goal is only contained in planning samples. Is there some mapping of sample uids to instructions?

Further, is there some way to reconstruct the full long-horizon episodes from the short medium-horizon snippets provided [here](https://console.cloud.google.com/storage/browser/gdm-robovqa/videos)? Consecutive file names do not seem to belong together.

Thank you very much in advance!

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.