Hi,
Firstly, thank you for your tool. Pretty well designed for annotation process ! Shortcuts are convenient even with a AZERTY keyboard :)
I would like to add new labels to a task already completed to match freshly new made tfrecords with new specification.
I succefully add one inside the admin panel. But I can not change his ID.
It's a trouble because the new label will not match with the other tfrecords dataset labelmap ids.
Have you any idea to solve this issue ?
Best regards,
--Selim
@selimbek, Hi let's clarify some details. Label IDs in the dumped tf_record files are generated by consistently during dump and don't depend on database IDs, only on order.
https://github.com/opencv/cvat/blob/0bb92f27c1f8a78cb66868f49b04b04f5ce32358/cvat/apps/annotation/tfrecord.py#L95
If you add a new label for the existing task, max_tf_record_id + 1 will be used for this label inside tf_record file.
Could you please describe what you want to get exactly? The same label map for the different tasks or smth else?
Yes exactly, I would like to share the same label map id's in tf_record files generated from different tasks. Even if the labels was added in different order.
Heyo,
I had the same issue with exporting TFRecords with differing label_map.pbtxt so I couldn't use this for training. I tried to change the "Raw" labels, but CVAT would not update my labels...
Solution:
It's a bit clumsy(and scary when you are overwriting 8 hours of annotating!), but it works.
Hope this helps any wayward annotators out there!
@dsgage
I did something similar to solve this issue:
Hi @selimbek @dsgage
I solve this issue similar with you guys 馃憤
I will close the issue for now, do not hesitate reopen it if you still have questions
Most helpful comment
@dsgage
I did something similar to solve this issue: