Mark unrelated neurons or multiple neurons in a single task

Feature Requests
#1

In some tasks, the amount of exposed neuron that is selected for the task is very minimal, leaving the solution to the task as mostly irrelevant.  But there are neighboring regions that look interesting.  I would like to mark these other “interesting” regions in a different color to submit multiple potential neural paths in a single task.

In certain other tasks, there are areas that are questionable, so I have experimented and found that neighboring regions don’t make sense.  It would be nice to mark these regions as irrelevant (a different color) to the task at hand.  It would be nice to mark these regions as not applicable to the assigned task.

One possible additional color to mark unrelated or irrelevant neurons would be the same shade of gray of the neighboring pixels.  I have noticed that when I look for regions that are similar, I try to focus on color differences, and having the unrelated neurons marked as gray would not draw my attention to the different areas.

1 Like
#2

Hi wingnut,


Thanks for the feedback.  One of the bugs that we are currently working on is the spawning of irrelevant tasks.  Once that’s fixed, all the tasks that you get should be more substantial.

In terms of marking things in different colors, I’m interested as to what your process is.  How would the ability to mark other neurons help you?
1 Like
#3

Hi! I don’t know if I understood wingnut correctly, but I can tell you my process when I am in doubt (maybe it’s the same as wingnut’s):
suppose in one slice I have an adjacent area that MAY belong to my neuron, but is not clear. I colour it, then I start “following” it on the other slices and see where does it take. At that point, I have some clues to say whether it really belongs to my neuron or not: - the global 3D shape, which may be strange (though this is kind of a feeling, I know!); - repeated contacts of the neuron with itself (meaning that in fact I started following another neuron. At this point, what I can do now is cancel out the new areas that proved to be wrong. But if I could colour them with RED, for example, that would remind me not to fall in the same trap, cause I already demonstrated that they are not my neuron! Then, if I make a mistake at another depth, I could end up with blue touching red, and I would know to be wrong!

Did all this make sense? (it’s 1:31 am here…!)

eluchinat
1 Like
#4

The ability of marking neighboring neurons would be beneficial to follow neurons that get smaller in size when you have to pay very close attention.

1 Like
#5

It would help in visualizing two different colors of what “is” part of the neuron and what the person knows isn’t part of it. Personally I will click on a part that I’m guess is part of the task, if it looks wrong i deselect it, but what is preventing me from clicking the same wrong branch on a different slide(same task) this “second” collor will help show you what is part of the “task Branch.” it could even help with finding the full branch( process of elemination)

      -Deadxdying

1 Like
#6

I work mostly from a laptop with a small screen.  I have to be efficient with the screen real estate, so have to pan and zoom in and out quite a bit.  On the desktop, it’s fine because I’ve got dual monitors and can move the 3D screen to one monitor while looking at the 2D slide.  On the laptop, I find the edges of the neuron that are certain (for example, at the beginning of a task, many neurons near the area of interest are clearly sectioned off.  If they can be marked as not connected the contrast will be higher).  So I flip back and forth between slides looking for differences.  When the shades of gray are close enough that it looks like a portion of the neuron is unmarked, I mark it.  However, when there is uncertainty, I flip through more slides to determine if it’s just a gap that will eventually be filled by my neuron or there is another neuron invading.  When it is an invading neuron, I want to mark that as “not mine” and move on.

1 Like
#7

I really like this idea. I find that synapses can be especially confusing, since several cells may be closely interleaved and the axons don’t always follow straight lines. Being able to mark adjacent cells in a different color could really help deciding which cell ambiguous or noisy regions belong to.

A contrasting color in the 2D view (say, red) would be useful. The other cells could either be ignored in the 3D view or be displayed with some transparency.

1 Like
#8


A contrasting color in the 2D view (say, red) would be useful. The other cells could either be ignored in the 3D view or be displayed with some transparency.

I agree completely. Also, the double colors would also help the developers easily identify synapses 

1 Like
#9

Yeah! It will be nice, if I will can fill the cube totally by different colors for each neyron!

Like in http://blog.eyewire.org/gallery/image-gallery/nggallery/image/berger-cube_dendrites-1/
1 Like
#10

It’s Here! http://blog.eyewire.org/introducing-explore-mode/

1 Like
#11

Fabulous! I’m using explore mode on nearly every cell I trace now.

Automatically committing exploratory areas when the “I’m Finished” button is pressed seems dangerous. Wouldn’t it be better to not accept cells with exploratory markings? Just add a dialog telling the user to either commit or delete exploratory neurons before submitting.

1 Like
#12

I’ve just finished an A.S. Degree in programming, but admittedly I know very little about AI. I actually used EyeWire as the introduction for my honors program capstone project, but I hadn’t really delved deep into it until recently. 


It seems to me that a lot of the time my decisions on whether or not a chunk belongs to the current neuron branch is based on whether or not it appears to fit better with a different branch, which is often difficult to determine. Of course, highlighting every neuron branch in a cube with a different color would just look to a human like a cluttered mass of color, but to an AI perhaps this could be a different perspective - matching chunks based not on how well they fit with the selected branch, but on which branch they fit best.

And perhaps this information could somehow be displayed to the players as well; maybe as semi-transparent red volumes indicating other nearby branches, and allow players to assign chunks to those so that they won’t accidentally add them to the solution? There could even be a mode where rather than tracing a single branch through multiple cubes, players identify all of the different branches in a single cube - marking each with a different color, with the ability to selectively hide/unhide each branch.

I can’t be sure exactly how difficult that would be for you to implement, or how memory- or processing-intensive it would be, but I assume you’ll want to map those other branches eventually anyway; getting them out of the way now would help to clear up some visual clutter and improve accuracy.

On an unrelated personal note, I’d love a glance at some of the source code, but I’m sure there’s not much chance of that happening.
1 Like
#13

Hey Hesulan, 


We actually have something similar to what you’re asking for.  http://blog.eyewire.org/introducing-explore-mode/  It’s called Explore Mode and it lets you color different neurons in a second color to help you tell whether or not they belong.  Check it out and let me know what you think.  Also very cool that you used us for your capstone project!
1 Like