Feedback on UI and improving productivity
-
by
msky
Dear Members of the Spacewarps Team,
After a few weeks of classifying lenses here is some feedback on how I believe my productivity could improve. In general my throughput is limited by the number of times the full image needs to be loaded and the related zoom/pan mouse actions. Tweaking the user interface in the following ways would widen that bottleneck:
- adaptive scaling of brightness levels in standard view
- zoom in on current mouse position (rather than the image centre)
- steeper zoom and faster panning
about 1) Obviously perception is very subjective. The point is that above a certain brightness level I need to zoom in every time, because potential features are simply outshined. It might be worth trying some non-linear mapping of brightness to pixel values (or histogram equalization) in the standard view. That should cut down on the number of tedious zoom/pan/brightness adjust action sequences, which leads directly to points 2 and 3...
about 2) Unless the suspect object is right in the middle of the image one always needs to zoom and then pan in the direction where the region of interest just vanished beyond the edge of the FOV. It would speed things up a lot to zoom in/out on the mouse location.
about 3) A few discrete zoom levels would suffice. As a workaround I keep notching up the sensitivity of the mouse wheel and also the lateral acceleration for panning - with limited success. Sure, that is somewhat device specific, but it is about thousands of more benign (wrist/finger) movements over the course of a classification session.
Somewhat related is a remark on simulated lenses: People might be more inclined to properly classify the (dreaded but necessary) simulations, which is what we do 99.x % of the time, when it does not involve zooming in. Waiting for the enlarged image is disruptive - certainly with my network connection at home.
Lastly, after having switched from my default Firefox to Chrome for the specific task at hand I was generally very pleased with the overall "classification experience" and robustness of the UI.
All in all, great work by the teams involved!
Posted
-
by
anupreeta
scientist, admin
hi,
Thanks for your feedback. I will let someone else answer on rest of your points but let me answer the following:point no. 2- a better approach would be to pan and bring the object of interest to the center and then zoom in..
if I've understood your remark on simulated lenses, then the point is that the images with simulated lenses have other objects that need to be classified too. Changing the image size or showing the image with a different zoom factor will not help in this case. Please let me know if you were raising a different point here.
Glad to know that Chrome works well !
Posted
-
by
msky
in response to anupreeta's comment.
Right, panning before zooming is clearly a better strategy and takes care of no. 2. Thanks.
The point about the simulated lenses was a different one: Arcs and quads are often so close that only one can be marked without zooming in. It is not clear to me whether the system actually needs more than one marker, however, the popup dialogues frequently contain text asking for it.
Posted
-
by
anupreeta
scientist, admin
in response to msky's comment.
Ok. What to mark is given in the FAQ. Please have a look at the FAQ since it may answer any other questions you may have, in general. Short answer - marking any one of the images is crucial, marking additional images is desirable.
Posted
-
by
bernmeister
My experience of productivity has highlighted another bottleneck. Going back to check on an image that may have been dismissed too briefly. It can always be picked up in the profile but a "back" button as easy to use as the "next" button would be great. Easier said than done, I know. Perhaps a function for future enhancements.
Posted
-
by
anupreeta
scientist, admin
Not adding a back button during classification was a choice made to keep the users focussed on classifying current and future images. I'd have liked a back button as well but "speed" is also important for this project since our aim is to get a large no. of images classified.
Posted
-
by
msky
A follow-up to my opening post:
The assumption that loading additional pixel data and the respective network latency is responsible for the delay when switching to the quick dashboard was premature. It is the CPU usage that spikes for a few seconds before the dashboard becomes available. Nonetheless, here are two ways of reducing quick dashboard access:
- using the browser's zoom capability (often mapped to CTRL -/+ as mentioned elsewhere on this board)
- Quads or arclets which are close but not too close can be flagged without zoom by first putting the (red) markers anywhere on the image and then dragging each to its intended location. Overlapping markers does not work when the separation is a small fraction of their size.
Posted