You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Cvat workspace is managed via code including python cvat sdk. (Ideally, the same actions shall be possible with the cvat web-ui for small projects). That includes task creation and annotation data export. Source data as mp4 files. More data, projects and cvat organizations than feasible to handle via cvat web-ui. Cvat v. 2.10.2
Problem
The annotation data export includes information about the frame numbers. Not associating individual frames with timestamps at cvat task creation creates complications and risks in the data pipeline down the line.
I could not find a suitable metadata class for individual frames (i.e. not that) or such that I can write to. Also no object, field, etc. that I could use to write said information to the cvat host.
I did not find a possibility to pass additional data, e.g. a list of timestamps, to any function related to task creation and data upload.
Export frames from mp4-files to pngs with timestamp in filename. Those names are later found as the 'frame_name' in an export to Cvat for Images or Datumaro or other. This solution is not favourable however.
Not ideal but still the most likely candidate so far: Set (automatically) image-tag (e.g. called frame_metadata) for each frame with a timestamp field in the attributes. The tags would be set using 'task.update_annotations(label_request: PatchedLabeledDataRequest)'.
Questions
Is there a way to hide a label or at least make it non-writeable in the web-ui? (For solution 4.)
Is there a another, cleaner way to write and retrieve custom, frame-wise metadata?
Can I upload and retrieve an additional file, e.g. a csv-file with the timestamps?
PS: I wrote initially, that ideally the same results shall be achievable using cvat_sdk and the web-ui, but I guess this will not be possible either way.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
Hi everyone,
perhaps you have come across this issue.
Premise
Cvat workspace is managed via code including python cvat sdk. (Ideally, the same actions shall be possible with the cvat web-ui for small projects). That includes task creation and annotation data export. Source data as mp4 files. More data, projects and cvat organizations than feasible to handle via cvat web-ui. Cvat v. 2.10.2
Problem
The annotation data export includes information about the frame numbers. Not associating individual frames with timestamps at cvat task creation creates complications and risks in the data pipeline down the line.
Current state
Task creation:
client.tasks.api.create(task_write_request = <my_task_write_request>)
task.upload_data( resource_type= ResourceType.LOCAL, resources= [<single_mp4_filepath>], params = <my_data_request_config>.to_dict(), wait_for_completion=wait_for_completion )
Annotation data export:
client.api_client.tasks_api.retrieve_annotations(id=<my_task_id>)
Reviewed Solutions
Questions
Is there a way to hide a label or at least make it non-writeable in the web-ui? (For solution 4.)
Is there a another, cleaner way to write and retrieve custom, frame-wise metadata?
Can I upload and retrieve an additional file, e.g. a csv-file with the timestamps?
PS: I wrote initially, that ideally the same results shall be achievable using cvat_sdk and the web-ui, but I guess this will not be possible either way.
I hope, somebody has some input. Thanks!
Beta Was this translation helpful? Give feedback.
All reactions