Other method of working in Eyewire


Has an investigation ever been done in a different method of working with AI?

Such as adjusting the grayvalue detection of AI?

Namely starting to color everything from a dark gray value e.g. 20% and then decrease until you see a few connecting neurons in 3D.

It can then be removed by hand.

After all, I work with an erase method instead of an add method.
Namely by coloring everything around the neuron, I see the structure better and I can erase the strange neuron parts while retaining the good ones.

If we can color all weak discolorations in this way, we can erase the bad ones and retain all the good ones without having to color in the small missed parts piece by piece.

Hi ggeu,

Yes! Seung Lab has been working on better AI for growing neurons. Improved cell tissue imaging methods with higher definition has aided us in creating better AI for other datasets. You can see some of this AI in action via the Mystic gameplay on our Zfish (zebrafish) dataset. This video clip shows the AI growing a cell in realtime and joining segments together: https://youtu.be/2t5ETAS0TMA?t=186

Of course, you can watch the whole video describing this form of gameplay and read more about Mystics and the Zfish dataset here: https://blog.eyewire.org/introducing-eyewire-mystic/

We have more AI projects in development including Neo which has an improved AI: https://blog.eyewire.org/2019-neo-update/