Help proofreading and working with FlyWire data

Do you have a question about cell morphology, gnarly slides in the dataset, or how to properly execute a split or merge?

This is the place to ask your questions! Please also see the Glossary of fly ultrastructure and morphology and FlyWire: Column Countdown blog posts as starting resources.

Creating a link using the “Share” button in FlyWire and posting it along with your question may also help you get a quick response.

I’ve been playing with tools and edits in the sandbox and for the most part I’ve got a hold of things I think, but I’m running into two issues.

The first is when trying to split segments, either by simple or multicut. Sometimes I can get them to work, quite frequently, I get this message:

Split preview failed: [400]: All supervoxel must belong to the same object. Already split?

Here’s an example of one area I hit this message with:

The other issue I have is when merging - quite often only a small segment of the original cell merges, while the rest completely disappears. In the below example, I’ve merged a small segment to the soma, causing the entire branching structure to disappear (I have also had similar happen in reverse - adding a small branch along a dendrite loses the surrounding dendrites and the founding soma, etc):

Thanks in advance for the help - I know it’s the weekend when I’m posting, just trying to catch on with these new quirks!


I experienced the same error message as Jaime as well. Here are two links where this occured:

Do note, that I was able to successfully perform the split in both links once. If I remember correctly, after the second split was performed, the cells in 3D disappeared completely, and I couldn’t get them to reappear again even with a normal page reload. When reloading the page with CTRL+F5 (loading without cache), everything appeared as normal, but the aforementioned error messages now show.

EDIT: Apart from the 3D render disappearing this link has had the same happened to it. I assume this is an issue with sharing a link and then performing the action that that link portrays. FlyWire


Same for me. I often experience the " Split preview failed: [400]: All supervoxel must belong to the same object. Already split? error.
I presume the AI just do not find where to cut.

I also have problems with cut/multicut (not tried merger for the moment) and cell disappearing totally.
Even a F5 do not resolve.

Like here : FlyWire

Impossible for the AI to cut. As single or multi cut, it can’t find where to cut and i got the “Split preview failed: [400]” message

1 Like

Hi! I just posted a blog post on common error messages.

I’ve gathered the info in this post from various resources, so hopefully it’s all correct, but it’s possible some info may be out of date.

If you find that any of the solutions in the post do not resolve your issues, or you have other error messages that are not listed, feel free to post here and we will look into them further. It’s helpful if you post the exact error message received and what the circumstances were when you received it.


Thank you for posting your questions here!

^^ Seconding Celia’s link to the blog post with common errors ^^

For most the issues posted above, you were most likely editing a neuron that wasn’t up-to-date, a safe practice is to deselect and reselect the neuron in the 2D. When you reselect the neuron, you’ll get the most recent version of that segment.


Thanks for the clarification on error codes! That’s very helpful.

I’m guessing that some of these errors have to do with the sandbox data set itself? For instance if I reload the default sandbox, everything is at a base state, but as many edits have been made, reselecting a neuron will load pretty vastly different neuron structures than the default load. In short it seems like the sandbox view loads ‘old’ data as default, but edits have since happened, so we’re not working with the same data as the default view anymore.

Given it’s the sandbox I’m sure that’s fine, especially since any links we generate for our “production exam” will contain the state we’re looking in, but I suppose it will be extra important for us to always be checking the ‘current state’ when we’re working in production!


Here’s a new one! There’s a chunk of unconnected neuron attached here that appears to be floating unconnected in both 2D and 3D. Attempting to preview this split results in the error “Split preview failed: [400]: Sinks and sources are not connected through the local graph. Please try a different set of vertices to perform the mincut.”

How would this segment be handled?

Given that, evidently, two parts (let’s call them R and B) which aren’t connected can’t be split from one another, I would try and find a connecting segment (S) that touches both R and B. S would then be merged with R and B, forming one big segment containing R, B, and S. I’d then perform a split to separate S from B, resulting in one segment containing B, and one segment containing R and S. Another split to separate R from S would complete the operation. In the case of your link, maybe something like this would do the trick:

1 Like

It looks like you’re getting a couple segments that were merged together at a strange place. TBH I have no idea why these segments were merged like this, and I think it would be rare to see something like this in the production dataset. You could still get segments that aren’t merged together at the right junction however.

For this case I used “find path” to determine where these two pieces were connected, and then placed some multicut points around the connection bridge shown by the path. It seems that it will result in a successful split.


Yeah this was a definitely weird thing to find, but I was curious how to deal with it! I hadn’t been able to find a working split but you did, thank you - very useful learning tool here!

In the sandbox, for the “Accessing the Production Dataset” i need more info about the “3 links showing places where an accidental split occurred”.
Do i have to look for two pieces already visible in the sandbox ? or do i need to find where there’s a missing extension, adding it and reporting it as two pieces to merge ?

You need to find a missing extension, add it (but don’t merge), and then share the link. As an example, here’s one that I submitted: FlyWire

EDIT: Corrected link, I am a derp!

1 Like

Your link just redirect me to the basic overview : - Nimbus Capture
Are coordinate correct ? 158581, 72226, 2189
But i understand, i just need to add extention whithout merging it. And report. Thank you for the answer bl4ckscor3 !

1 Like

Thank you, that’s exactly what i was asking. And it answer perfectly !

This is just a question of curiosity: What is this structure with a high density of black clusters?

Link to data: FlyWire

1 Like

I’ve asked a few folks in our Princeton lab, and even they aren’t sure! It looks like some overstaining inside a glial cell to me (but I may be wrong!). I’ll see if any other fly researchers know :face_with_monocle:

1 Like

What type of cell is this? My best guess was an MI type.

Yes, I agree that’s it’s most likely a MI type cell (maybe Mi3) especially since it terminates/arborizes in the Medulla only.