Image Object Detection UI Feature Request

Hey all, I’m back with a new post, it’s been a while!

I’ve done a fair amount of labelling now via your object detection labelling screen, and I like the way it works. However, there are a couple of things missing that I’d really like on there.

  1. A way of turning off the built in method of attempting to detect where objects are in an image. Sometimes it’s accurate, but other times it’s not, and when I’ve got a minimum of half a dozen or so object frames left over it can be time consuming to remove them or reposition them. So a way of turning off this feature would be great (if it already exists, could you tell me how please?) That or a button to remove all frames left over from the previous image labelling.

  2. Some clearer way of telling the user that the items labelled outside of the square on landscape 16:9 images aren’t processed would be great; they aren’t processed right? At the moment it’s faded, but that can be misleading, totally blanking the area out, increasing the opacity of the dark screen or putting horizontal lines across the dark screened area might be a good idea. Or even a bit of text at the top saying “Only the items within the fully visible square will be assessed”.

My priority would be number 1 of the two above things, as it would saves me a 10-20 seconds per image, and when I’m working on hundreds of images, that helps massively!

Cheers everyone, hope you’re doing well!

Replying to my post here, but I just saw the drop down allowing you to use the “Classify using YOLOv5” option and WOW. So much better, so I guess that answers my own question partially. I still think having a way of turning the auto classification off entirely would be good though.

Very good feedback @TechDevTom ,

I’m creating an internal feature request.
About the area of labeling it is tricky because it will depend on the resize method you will use. (Squash, fit to shortest axis, fit to longest axis).
But I do agree it is not super intuitive.

I’ll try to check with our Core Engineers / UX designers if they can come up a better way to guide the user.



1 Like

Cheers @louis, much appreciated!