Only pose information loaded when choosing to download all the annotations of egoexo4d

[likanchuan09171]

Hello,

I run a command like 'egoexo -o /data/ --parts annotations metadata captures expert_commentary --uids 83ab91c7-ba47-45be-bd19-38d9246baae4 -y', but only annotations/ego_pose is downloaded, while there are no other annotations e.g. narrations and expert_commentary.

I believe that these parts of this take exist , as in the visualizing website: 'https://visualize.ego4d-data.org/iiith_cooking_11_1?v=%22EgoExo+v1%22&p=3&s=%2283ab91c7-ba47-45be-bd19-38d9246baae4%22'.

Did I misunderstand the usage or is something wrong with the CLI downloader?

Can you help me to solve this?

Thanks,

LKC

[miguelmartin75]

Yes this may be a bug, I will look into this.

To unblock yourself, are you able to download those parts without the filter? They should not take much disk space.

[likanchuan09171]

I have tried commands filtered by splits like 'egoexo -o /data --parts annota tions --splits val -y'. It ends with a error as below

This time there are some annotation files left like 'atomic_descriptions_val.json.F14eE010' 'expert_commentary_train.val.73aB0F71'. Manually converting filenames to json results in meaningful content, but some of their lines are incomplete and therefore do not constitute a proper json format.

Next I'll try if these annotations could be downloaded fully without any filters.

[miguelmartin75]

Those are partial files when downloading from S3, i.e. download did not complete (e.g. interrupted). Use -d on the CLI to delete auxiliary files (be careful with this option and avoid using filters)

[SJTUwxz]

Closing this issue for now. Please feel free to re-open it if the problem persists. Thank you!