Open hohonuuli opened 2 years ago
My proposal is to drop the use of project completely. There's a column, observations.observation_group
that should be used to track this. Currently, the way it's used is that primary annotations, those created by the video lab, are assigned to a group like ROV
or i2MAP
that groups them by platform type.
The ML annotations were imported into a group named ROV:training-set
. These annotations might be redundant, essentially double counting observations. By putting them into a special group, they can be excluded from queries for quantitiative/qualitative data. Note that the other common group for ML annotations would be i2MAP:training-set
.
We met with the video lab today and discussed this. The current thinking is to keep an observer
and verifier
field in the associations's link_value
JSON.
The vars-localize and vars-gridview tools add additional information to an association's
link_value
json used to define the bounding box. These fields are:observer
- Who drew the boundingbox (using vars-localize)verifier
- Who confirmed the id of the bounding box (using vars-gridview)project
- info about what project the localizations were created for. e.g. FathomNetWhen I merged the ML generated localizations into MBARI's main VARS database, I dropped these fields. So this issue is to track a discussion on how best to formalize the tracking of the data in these fields. I'd prefer to remove them from the JSON as it creates a lot of noise, but that might not be the best course of action. This should be a discussion with @kevinsbarnard and MBARI's video lab.