I find the current version of Project Discovery pretty annoying to do because it takes perhaps a fraction of a second to spot the clusters but then half a minute of tedious clicking to transfer that information to the computer. If the UI worked a little differently, we could do it with a lot less clicking.
First some background. The tutorial suggests that we draw the polygons tightly around the clusters like this:
But this is unnecessary as the scoring system doesn’t care if we include extra empty space around the polygon. In the livestream it was also confirmed that this is Ok. So we can do our polygons with less clicking:
And still get a perfect result:
So I propose an UI that does the same thing but with even less clicking. When a task starts, there would be a ready made fixed border surrounding the whole area:
User would start drawing by clicking some point on this border:
And then clicks another point on the border to end the line he is drawig:
And that’s it. Done. Ready to click ‘Submit’ after just two mouse clicks instead of ten like in the existing UI.
If a straight line won’t do, we could click intermediate points along the way before clicking the border again. If there are more than two clusters, we would add more dividing lines that can start and end on the border or on any previously drawn line. But no looping back to end the line on itself. If there is just one cluster, no drawing is needed at all as the initial border covers it, so we would just click submit like in the ‘no transits’ cases in the exoplanet version.
Line endpoints and intermediate nodes can be adjusted by dragging but the nodes on the border can only be dragged along the border, not off of it.
Drawing a line that crosses another line would automatically add a node to the crossing point. So if we need to divide the area to four squares, we could do it with just two crossing lines from edge to edge.
This subdivided plane is then easy to convert to polygons if the site where the data gets uploaded requires it. But this is work that computers are good at, so humans should not be bothered with it. Each polygon could then even be reduced to the convex hull of the cells inside it to produce polygons that look like those ‘gold standard’ polygons. But again this is work for computers, not humans.
And some further tweaks:
There should be some ‘magnetism’ in the lines so that clicking near one is sufficient to snap the new node on the line. And on the nodes too so that creating a new node near an existing one would reuse the existing node. Wider tolerance than in the current UI where closing the polygon at the start point is hard to do on a high resolution screen. Part of this difficulty comes from the ambiguous mouse pointer where the active point won’t match the expectation.
Also there should be a way to remove just the last point added after a misclick. For example by pressing the backspace key.