annotate/browse mode: labels with spaces and parts

[dobkeratops]

1. Great to see some of the recent labels enabled (street light, fork lift etc) but I was trying to enter these in the browser mode: it doesn't seem to find any examples; I wondered if it's not correctly parsing labels with a space in them (I tried using an underscore , or trailing space/no trailing space, seeing what the autocomplete did.. I did see these new labels in the autocomplete; I do see them in 'add labels', an other labels like sofa,table did get enabled fine)

2. should it be able to find parts e.g. "head/person" etc - it doesn't seem to , but I do see things like "dog.has=head" in that dropdown list (does it look for a different syntax to identify parts here?)

[bbernhard]

Many thanks for raising that issue!

I just checked directly in the database...there are a few labels where all the images that use the label are either not unlocked yet or are currently in "quarantine" (images where we need to blur out faces)

But you are also totally right with your assumption: there is indeed an error in the parser (the parser naively interprets the space as a label separator). But I am currently working on that part of the code (in order to implement #157)...maybe I can kill two birds with one stone here :)

regarding 2: yeah, the parts have a slightly different query syntax at the moment...The first implementation of the parts concept dates back to the early days of ImageMonkey. At that time I thought that a more natural representation of sub(-parts) might be a good idea. Looking back on that decision...I am not sure anymore if it was a wise one. I think it might be a good idea to change that at some point...or at least also support the "slash-notation".

[bbernhard]

(Thinking again about it, I still think it might be a good idea to let users work on annotation tasks where either the label and/or the image is not yet unlocked. While going through the labels and the images for unlocking, I haven't found a lot of labels/images that we absolutely can't use. Most of the time it's just a person that needs to be blurred out and everything is fine. My hope is, that if we let the image owners operate on the images even when they are not unlocked yet, that we get >= 90% success rate (i.e we can directly unlock images + completed annotation tasks without any change). For the remaining 10% we need to do slight adjustments (e.q blur out faces)..but if we keep the image dimensions the same we could still keep the annotations that were done on the original image.)

[dobkeratops]

ok regrind the slash annotation etc.. no need to rush any solutions; there's plenty to get on with elsewhere

[bbernhard]

@dobkeratops labels with spaces should work now

[dobkeratops]

confirmed , awesome