feature request - bulk operations

[N. Hawkins]

It would sure be handy to have some bulk operations available to make the annotations process more efficient.

I've been using the Megadetector model to help weed out blank images. It's been pretty good. Some deployments it's been really good. Others a little less so. Very few false positives for the "blank" classification. False negatives vary. For this, it's been useful to use it.

However, it's REALLY BAD about the "Vehicle" tag. With my study protocol, vehicles are an impossibility. I think the problem is that it recognizes the circular shape of the culverts in all of my images and marks a bunch of them as vehicles (mistaking them for wheels?). At any rate, I know that if something is marked as a vehicle, it's wrong. I would like to be able to select all images with the vehicle annotation and remove that annotation.

So far, all images marked as containing a "Vehicle" have actually been blank. And weirdly, they've almost entirely been nighttime images. I recognize that adding annotations in bulk is problematic for a given study. But the ability to remove annotations in bulk would be handy. At least that way, I wouldn't have to click "Unmark" on all of the hundreds or thousands of incorrectly annotated images before then selecting the correct one.

I'm all about being more efficient with my time. Shoot, even having the option when creating a deployment to tell Agouti that "Vehicle" is an invalid annotation that is not to be used would be great.

Nate

[agouti]

Dear Nathan,

 

It’s a little hidden right now, but you can bulk delete observation directly in the observations table. This includes vehicle annotations if you set the filters right. Try this:

 

 

If you make a very large selection, it may take a while to process everything. Hope this helps!

 

Best,

Yorick

--
You received this message because you are subscribed to the Google Groups "agouti-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to agouti-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/agouti-users/1746b73c-e211-428d-be8d-f21526aee165n%40googlegroups.com.

[N. Hawkins]

I did encounter this possibility. Thing is, that's not quite what I'm after. I just want to remove the annotation. Not the images/observations themselves. In my case, none of the vehicle annotations are correct. They're almost entirely blank, but only almost. A few detections sneak through the AI processing. So I do need to go through everything annotated by the AI as blank, also, to make sure there's not something important in there. The vast majority of those are correctly annotated by the AI so I only need to do occasional corrections manually.

In my project, that's mostly herps. Snakes and salamanders mostly, but also the occasional box turtle or toad. So I definitely don't want to bulk delete anything the AI annotates as a vehicle. I just want to improve the efficiency of how I correct the annotations. As an example, I have a few deployments I haven't yet completed processing and there's several thousand images annotated by the AI as vehicles (Megadetector v6 definitely does a better job with this, but for some reason or another, that model is not working right now and I have only been able to successfully use v5a). Going through each one and then "Unmarking" it in order to then manually annotate it correctly adds a lot of time to my work over the course of all of these. So the ability to "Unmark" them in a single bulk operation would help a TON. 

It would be a bit more efficient for me if I could even change the annotations from "Vehicle" to "Blank" because it would reduce the clicking I'd have to do. I could do a quick flip through the images to verify they're blank and only correct what slips through.

So for now, my only option is to manually "Unmark" each one individually. It just takes a LOT of time. More time than flipping through all the images in file manager and then manually entering the relevant data into a spreadsheet. The only reason I'm sticking with using Agouti right now is because I like how it organizes things. But there still some things that would help to make Agouti more efficient to use than it currently is.

Nate

[Blair George]

If you bulk delete the 'vehicle' observations per Yorick, it will not delete the underlying images/sequences.

After bulk deleting the observations, you can go back to the Deployments screen and click to 'Annotate' each deployment. This interface will automatically take you to the first un-annotated sequence of that deployment, so you can review and either click the 'Blank' button or add your observation. Then there will be a 'Next unannotated' button at the top so you can skip directly to the next unannotated sequence of that deployment.

You have to repeat this for all deployments that now have unannotated sequences, but you can use the 'Hide 100%' option to filter for just those deployments.

We have found this to be the most efficient way within Agouti if you have hundreds of observations or more to review. While the observations interface is powerful at filtering a subset of observations, there is not a good way to bulk-annotate just that subset you are interested in.