User interface quicker. Too many clicks per annotation.

[alexdominguez09]

Hi, I am using this tools to distribute it to students so they can annotate images. We have been trying it for a while, in UBUNTU 16.04 machines and it is a very simple and straight forward tool. Perfect for the job.

However, our students claim that they need too many mouse clicks per image, and after evaluating the user interface and usability, we have come to the conclusion that the label image tool would need an improvement in the usability.

At the moment, a whole annotation requires:

Next Image button (1 click) -> Create Box button (1 click)-> actual box creation (2 clicks) -> Select label dialogue (2 fast clicks) -> Save button (1 click) -> Enter (1 touch key) ->

Next Image ... -> ...

That is 7 clicks plus 1 enter key.

Our request (as we dont have the knowledge or resources to accomplish it) would be to reduce it to 2 clicks.

That would be, once the image to work with is on the screen:

• 1st click: Selecting the first corner of bounding box (xmin, ymin)
• 2nd click: Selecting the second corner of bounding box (xmax, ymax), select automatically the label (the previous selected one), save it with the recommended name in the previous location(name of image.xml) with no dialogue box, then go to next image and activate the Create box tool, signaling the image as a non edited one (not to confuse the operator), as after half an hour, the operator start to get confused with the images, so it would be perfect a visual signal to see if the image is being edited.

That way, we reduce the work form 8 interactions to 2. Obviously, that would be implemented with a mode that user could select (turbo mode), so the user can always go back to the normal mode, to change the label, location..or any other setting.

Well, that is our result of a studding after using it for a while.

Thanks for the good work.

Alex

[alexdominguez09]

I forgot to comment, that our proposition would be such, as it is faster to annotate just one label per operator, than going to several labels per image. Eg. One operator labels only dog, other operator label cats, other does cars, ..and so on.

[tzutalin]

Hi @alexdominguez09 , Thanks for your suggestion. I will take your suggestion into account. If adding a turbo mode, it will be better. I even thought I should create a tool to help people label automatically for the first time. I will try to think about it and do it when I have free time.

[zuphilip]

Usually, it can also be that there are more object of the same label you have to annotate. Thus, I am not sure how often we can automatically continue after the first box is entered. Moreover, it might be that I want to edit the box a little after the first sketch, but then with your turbo mode I would already be on the next image.

I like the option that one does not have to choose a label, if there is a default label which is used all over. This would be similar to https://github.com/tzutalin/labelImg/issues/49 .

I envision more, that creating a new rectangle and the next page will be entered with the keyboard shortcut. Thus my left hand is on the keyboard pressing a, d, w and my right hand entering the rectangles with the mouse. The annotations are IMO automatically saved when I click on "next image".

[cooli7wa]

I also think two click is nice, one click one corner.

But I think dialogue box is necessary, because:

1. sometimes need to add new class.
2. we have more than one classes, and we don't want to modify it after auto choose. but don't need save button, just click the right class once.

So I modify this in my fork, you can try it.

[anxxFURY]

What are number of clicks.. Can someone plss explain, we have a course project rgarding it..