Clarification on Dataset; Motion Prediction

[arbaazkhan2]

Can someone illuminate what the main difference between the scenario/ folder and the tf_example/ folder is in terms of the actual data? (also what are the subfolders? for example, what is the difference between testing and testing interactive?)

I was able to run the tutorial using a random tfrecrod file from tf_example/training folder. But when I picked a random file from scenario/training (uncompressed_scenario_training_training.tfrecord-00000-of-01000) it was missing roadgraph_samples/dir (tensorflow.python.framework.errors_impl.InvalidArgumentError: Feature: roadgraph_samples/dir (data type: float) is required but could not be found. [Op:ParseExampleV2])

Any reason if this is a bug or if the two folders are different from each other based on something?

[s-ettinger]

Hi Arbaaz, The difference between the two folders is the format of the data. They generally contain the same data, but the tf_example folder contains tfRecord files of tf.Example protocol buffers while the scenario folder contains tfRecord files containing Scenario protocol buffers defined here : https://github.com/waymo-research/waymo-open-dataset/blob/master/waymo_open_dataset/protos/scenario.proto. The tf.Example protos are generally the same information as the Scenario protos but converted to tensor form for easier use in modeling (the tutorial only works with tf.Example format data). Because the tf.Example data has fixed sized tensors, some data may be truncated from the original in the Scenario proto. See the tf.Example specification here : https://waymo.com/open/data/motion/tfexample

Please see the full description of the data here on the Waymo Open Dataset website: https://waymo.com/open/data/motion/

[nnsriram97]

Is it possible to infer lane centerlines for the ego car position on the perception dataset?

[s-ettinger]

Hi Sriram. I'm not sure I fully understand your question. Are you asking about using the map data from the motion dataset with the perception dataset? If so, those datasets are distinct. There is not a correspondence between the two datasets.

[nnsriram97]

Yup, that was my question. Thanks!

[zhangpur]

Hi, @scott-ettinger, What is the difference between xxx_interactive and xxx? I cannot find the descriptions in https://waymo.com/open/data/motion/. Thanks

[s-ettinger]

Hi, The interactive set provides pairs of interacting agents to be predicted while the standard set provides up to 8 objects to predict that are not necessarily interacting. The interactive set is used for the interactive challenge described here: https://waymo.com/open/challenges/2021/interaction-prediction/. The standard set is used for the motion challenge described here: https://waymo.com/open/challenges/2021/interaction-prediction/ . Please let me know if you have further questions.

[zhangpur]

Thanks :>.

On Thu, Jan 27, 2022 at 1:45 AM scott-ettinger @.***> wrote:

Hi, The interactive set provides pairs of interacting agents to be predicted while the standard set provides up to 8 objects to predict that are not necessarily interacting. The interactive set is used for the interactive challenge described here: https://waymo.com/open/challenges/2021/interaction-prediction/. The standard set is used for the motion challenge described here: https://waymo.com/open/challenges/2021/interaction-prediction/ . Please let me know if you have further questions.

— Reply to this email directly, view it on GitHub https://github.com/waymo-research/waymo-open-dataset/issues/295#issuecomment-1022438476, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGQRPHZYWUG6X56DT7MZS3TUYAXK3ANCNFSM44LHBLVA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

You are receiving this because you commented.Message ID: @.***>