You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I think this issue is already known, but wanted to clearly document it and figure out what the best path for doing this in the future will be.
My ultimate goal is to efficiently find how edits affect the level2 graph. This is currently not easy from the information provided in the edit logs for a couple of reasons.
First, for merges, only a single supervoxel edge that was added is stored.
Supervoxel IDs: [89385966883691061 89385966883698825]L2 IDs pre operation: [161443560921497602 161443560921497601]L2 IDs post operation: [161443560921498196 161443560921498196]
Now, can compare this to some brute force code I wrote to explicitly compare the level2 graph before and after an edit. It uses the bounds argument to make this much much faster than pulling the whole level2 graph.
In other words, there are many more edges (in the level2 graph) that were added than would be suggested by just mapping the added supervoxel edge to its level2 IDs before and after.
A second issue (and this one is perhaps more just me not knowing a good path for mapping these changes) - even though splits have multiple supervoxel edges written down:
i.e. every one of those supervoxel edges was just within the chunk, so everything that got removed doesn't really help us with mapping changes at the level2 graph. For comparison, this is the result I'm after:
So, in summary, the first issue is that even at the supervoxel-edge level, there's missing information about which supervoxels get changed. The second issue is that even when more edges are written down in the log (splits), that's not enough information to reconstruct the changes for the level2 graph. This is because one level2 node was replaced by two nodes in the split, each of which inherited some of the edges from the original. I am wondering if anyone can think of a way to recover what those edges are without my hacky solution of going via the level2 graph pre- and post-edit, or if that is really the only path?
The text was updated successfully, but these errors were encountered:
I think this issue is already known, but wanted to clearly document it and figure out what the best path for doing this in the future will be.
My ultimate goal is to efficiently find how edits affect the level2 graph. This is currently not easy from the information provided in the edit logs for a couple of reasons.
First, for merges, only a single supervoxel edge that was added is stored.
[[89385966883691061, 89385966883698825]]
For this operation, there is one supervoxel edge that is stored as "added".
We can map these modified nodes to their level2 ids at the time of the operation.
Now, can compare this to some brute force code I wrote to explicitly compare the level2 graph before and after an edit. It uses the
bounds
argument to make this much much faster than pulling the whole level2 graph.In other words, there are many more edges (in the level2 graph) that were added than would be suggested by just mapping the added supervoxel edge to its level2 IDs before and after.
A second issue (and this one is perhaps more just me not knowing a good path for mapping these changes) - even though splits have multiple supervoxel edges written down:
Most of the time, they are just going to be the within-chunk edges
i.e. every one of those supervoxel edges was just within the chunk, so everything that got removed doesn't really help us with mapping changes at the level2 graph. For comparison, this is the result I'm after:
So, in summary, the first issue is that even at the supervoxel-edge level, there's missing information about which supervoxels get changed. The second issue is that even when more edges are written down in the log (splits), that's not enough information to reconstruct the changes for the level2 graph. This is because one level2 node was replaced by two nodes in the split, each of which inherited some of the edges from the original. I am wondering if anyone can think of a way to recover what those edges are without my hacky solution of going via the level2 graph pre- and post-edit, or if that is really the only path?
The text was updated successfully, but these errors were encountered: