Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Downloading 'bridge' Dataset using tfds-nightly #25

Closed
Wolfwjs opened this issue Dec 1, 2023 · 7 comments
Closed

Downloading 'bridge' Dataset using tfds-nightly #25

Wolfwjs opened this issue Dec 1, 2023 · 7 comments

Comments

@Wolfwjs
Copy link

Wolfwjs commented Dec 1, 2023

Hello,

I am encountering an issue while trying to download the bridge dataset using TensorFlow Datasets (TFDS). I am currently using tfds-nightly version 4.9.3.dev202311230044.

Issue Description

When I attempt to load the 'bridge' dataset with the following code:

import tensorflow_datasets as tfds
tfds.load('bridge', data_dir="mypath/“)

I receive an error indicating that the dataset cannot be downloaded.

In addition to the issue described above, I am considering the possibility of manually downloading the dataset from the following URL:
https://rail.eecs.berkeley.edu/datasets/bridge_release/data/tfds/bridge_dataset/1.0.0/

Before I proceed with this approach, I have a few questions:

  1. Is manually downloading the dataset from this URL a recommended solution to the problem I am experiencing?

  2. Will manually downloaded data integrate seamlessly with the TensorFlow Datasets API, or are there additional steps required to ensure compatibility?

Any guidance or suggestions regarding manual downloading would be greatly appreciated.

Thank you!

@Wolfwjs Wolfwjs changed the title Issue with Downloading 'bridge' Dataset using tfds-nightly Downloading 'bridge' Dataset using tfds-nightly Dec 1, 2023
@kpertsch
Copy link

kpertsch commented Dec 1, 2023

Thanks for flagging this -- I was able to reproduce the error in our dataset Colab, not only for Bridge but for all datasets. I suspect this is an issue with the current tfds nighly. I tried with some non-robot datasets and those still seem to work -- @quanvuong could you check with the TFDS folks what's up?

image

In the meantime: for downloading the datasets you can use the "manual download" approach:
You can get the raw bucket path using the command from our colab:
image

Then you can download the dataset with gsutil -m cp -r dataset2path(dataset_name) <your_target_path>
-- sorry for the inconvenience!

@quanvuong
Copy link
Contributor

quanvuong commented Dec 1, 2023

Raised the issue to tfds team here tensorflow/datasets#5170

@Wolfwjs
Copy link
Author

Wolfwjs commented Dec 4, 2023

Hi @kpertsch , thank you for your suggestion. I would like to confirm, if I use the command gsutil -m cp -r gs://gresearch/robotics/bridge/0.1.0 /mypath/, does it download the latest bridge v2 data?

@kpertsch
Copy link

kpertsch commented Dec 4, 2023

Yes, this will download the Bridge v2 dataset that's part of OXE -- it is slightly differently formatted than the officially released Bridge dataset & lacks the additional camera streams. The official Bridge data is linked on the Bridge website and can be downloaded in TFDS format here: https://rail.eecs.berkeley.edu/datasets/bridge_release/data/tfds/bridge_dataset/1.0.0/

@Wolfwjs
Copy link
Author

Wolfwjs commented Dec 4, 2023

Perfect, the Bridge v2 dataset is exactly what I need. Additionally, could you please explain what is meant by "lacks the additional camera streams"? Does this refer to a lack of image or video data?

@kpertsch
Copy link

kpertsch commented Dec 4, 2023

Basically the dataset in OXE only has a single camera stream per demonstration (because RT-X was only trained with single camera input), but the official Bridge dataset from the website has up to 3 additional camera streams on some trajectories. We will update the Bridge data version in OXE with a future update to fully match the officially released one!

@Wolfwjs
Copy link
Author

Wolfwjs commented Dec 5, 2023

Thank you for your patient response! I look forward to your updates, and for now, I will start with a single camera stream.

@Wolfwjs Wolfwjs closed this as completed Dec 5, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants