cvat-ai / cvat

Annotate better with CVAT, the industry-leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale.
https://cvat.ai
MIT License
11.82k stars 2.89k forks source link

Meta information: Adding new fields to Track and Video models #1101

Open rcbop opened 4 years ago

rcbop commented 4 years ago

Hello,

I forked the repository and I'm currently trying to create a custom format by following the guide provided here but we currently need to store some additional information into the annotation object. Those fields are:

Track related fields:

Global job related fields:

Could I just add new fields for Track here into the TrackedShape model And those ones related to job here, Would I need to change anything else aside the format dump/load handlers?

Also any tips on how to add those fields in the React UI?

nmanovic commented 4 years ago

@rcbop , thanks for the question. I can advise only on a general solution which we will be able to merge into upstream.

You can workaround the problem (even I don't like the workaround it will help you complete the job quickly without any code customization)

It seems that we need some meta information for tasks/jobs/frames. I don't know yet how to implement it. But I see already several requests where it can be helpful.

rcbop commented 4 years ago

Sure thing, I will try to come up with some generic metadata storage solution for the annotation related db entities and get back to you. Thanks

rcbop commented 4 years ago

@nmanovic turns out we took the workaround path. Global job related fields were added as a dummy annotation points at the beginning of every video. Everything else became immutable attributes of the annotations. Those are treated by the parser to dump/load the annotations file. Thank you.