nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Image classification: Add index number to labelled objects list

RossSmail

If you're dealing with large numbers of labelled objects, it can get a little confusing determining where you are in the list, as there is no index number. If, for instance, you are going through objects deleting ones that you think are inaccurate, it is easy to 'lose your place'. Just having a column with the row number in would make the workflow easier.

Vind meer berichten getagd met

Status: Openen

Reacties

RossSmail

Hi Jeffrey,

Thanks for the detailed response, I will check out your suggestions, but would also say that my 'idea' (such as it was) was just to add an (or display the internal) index number to the grid showing your labelled objects. I've managed to get excellent results from the toolset, even when using my hooky workflow, so kudos to those involved, it's a great piece of functionality.

Cheers,

Ross

JeffreySwain

Hi Ross,

I apologize for the delay. So I was meaning something like this. Since the training manager allows you to create individual samples that are within the same class, then I meant you could 'group' or 'collapse' them as I have in this graphic.

After re-reading your statement, I don't think that is what you want to do. So I would propose this instead, perhaps a different way of doing this. I would recommend trying to take advantage of some different application. Since what you described is more of a observation, than an image classification training sample you could take you imagery (thinking it is an image or imagery service/layer) and then use something like Field Maps or Quick Capture to collect the data as you desire. You could publish your imagery data in ArcGIS Image for ArcGIS Online.

I know it is thinking a little outside the box, but if you used quick capture for example, you could make the 'types of animals you are observing.' Let's say duck, deer, rabbit, squirrel, bird for rough examples, then you could make a quick capture app with each of them as the 'observation' in a feature layer in ArcGIS Online. Then when you found a bird, marked it ion the image - then it would record it as a bird and then give it a number. You could then quickly spin thru the image and have the data you are looking for with numbers attached.

What you are doing with the training sample is not as designed. The multiple selections would be taken in a group to generated a Esri Classifier Definition, which would then be used to classify an entire image. From your description, the samples have to be found with intense scrutiny to assess what type of animal is there. So if you were to take your training samples and then create the .ecd, then I am not sure the classification is valid, since you said each of the observations are of different animals.

The training sample manager will allow you to quickly create a point file, but not in manner you are wanting I think.

Here is the info on Quick Capture and Field Maps. I think they maybe worth your time. I know each is a 'mobile' app, but on an Ipad or similar device you can load up the imagery layer you want and then push that through and zoom/classify at your leisure. Just be sure not to use the gps of the device you are using.

RossSmail

Hi Jeffrey,

Thanks for taking a look at this. I may well be doing something dumb - it wouldn't be the first time.

I am identifying a particular type of animal on a set of air photos which are stored in a mosaic. Each animal I create a training sample polygon for appears as a separate Labeled Object of the class selected when I create the polygon - ie, each line in the screengrab is a separate adult animal. I am going back through the animals I have captured to review them, having created a model and then used it to detect objects, and noting issues with the detection - and sometimes I find ones that I want to delete because I either want to recapture them or because I think they're causing issues in my model generation.

I don't understand what you mean by 'combine them from time to time'. I don't see any options for doing this, and I'm not sure how it would benefit me anyway, if I want to be able to drop training samples. But I guess I'm probably not understanding something - I have read the documentation, but perhaps I need to do so again...

Cheers,

Ross

JeffreySwain

Hi RossSmail,

Out of curiosity, can you describe the workflow you are using here? If they are the same class, then why not combine them from time to time? Were the training samples pre-generated or automated and require you to confirm if they are valid? I can definitely see where this could be hard to manage, but wanted to know more about your workflow for this process.

Please let me know.

Doe mee aan het gesprek

Meld je aan om te posten, content te volgen en meer. Nieuw hier? Registreer gratis.