Possible to pre-populate annotations?

[dopplesoldner]

Hi

I was wondering if its possible to pre-populate annotations into vatic. The motivation behind this is that we could use our existing object detection and tracking framework and only label the "missed objects". Thanks

[kr-ish]

I would like to know how to do this as well, thanks!

[thomaspreece]

I've looking into this myself and it appears that you can just dump the annotations into the sql database that vatic uses and they will pop up on the interface as if hand annotated. (This was only tested using offline mode, your mileage may vary)

[nileshgu]

@thomaspreece I guess this is what I need. Would you mind telling me how to dump the annotation into the SQL database. I have .txt file which is in same format as we get after dumping the "had annotated" data.

[kr-ish]

The SQL database doesn't seem to store annotation in the same format as the output files it generates - If I recall correctly, for each annotation, the database stores the annotation's state at the start and end of the video segment, and any changes (position, out of view, occluded, attribute change) in between.

So if you had an annotation that started at frame 50 and went out of view at frame 100, given the video segment to be 321 frames long, you would have (I think):

[[ x_min y_min x_max y_max frame lost occluded]] 1: x_min y_min x_max y_max 0 1 0 ... 2: x_min y_min x_max y_max 50 0 0 ... 3: x_min y_min x_max y_max 100 1 0 ... 4: x_min y_min x_max y_max 321 1 0 ...

You could probably write code that detects changes from line to line of the .txt and only injects those into the database.

Hope that helps (and is correct)!

[fucusy]

is there anyone have pre-populate annotations into vatic? I also want to do that. I annotate every frame of my video by deep network, but I need to correct it by vatic.

[Alcatraz0000]

Did someone find a solution?