Folks who would also like to see this dataset in tensorflow/datasets, please thumbs-up so the developers can know which requests to prioritize.
And if you'd like to contribute the dataset (thank you!), see our guide to adding a dataset.
I would like to contribute it.
Should I implement it for custom resolution or If fixed then what resolution should the dataset support, optimally?
The paper Self-Supervised Visual Planning with Temporal Skip Connections has used 64 x 64 frames only.
Thanks for your interest!
Looking at the paper code (https://github.com/febert/visual_mpc/blob/master/python_visual_mpc/video_prediction/read_tf_record.py), my first impression was that the original data supported up to 128x128. The documentation for bair_robot_pushing_small also sounds like frames at resolutions higher than 64x64 are in principle available, but I am not 100% sure this is actually true.
A data loader for custom resolution would be ideal but involve more effort. As a start, 128x128 - if possible - would be great and make visual details a lot more discernible.
I see here that the fake dataset for bair_robot_pushing_small was created for size 64 x 64.
So, should I just replace that data with 128 x 128 and then explicitly set the path of fakedata in the testing of one of the dataset builders. Or should I generate a separate fake dataset for the 128 x 128 version?
Maybe generate a separate one?
I have opened a PR. Can you please review it @martinengelcke ? https://github.com/tensorflow/datasets/pull/2279
What's the original resolution of the images in DATA_URL = "http://rail.eecs.berkeley.edu/datasets/bair_robot_pushing_dataset_v0.tar"?
Looks like its 512 x 640.
Closing this issue. See #2279