Do I have to use Google Cloud to use pychunkedgraph

[liuyx599]

Assuming I have an EM image and segmentation of an electron microscope dataset, my previous approach was to generate corresponding mesh and skeleton using igneous. Afterwards, I will use the Nodejs server so that I can access these generated meshes and skeletons locally and visualize them through the neuroglancer.
If I want to use pychunked graph to store and mount them on the server now, must I use Google Cloud?

[liuyx599]

Alternatively, let's say I now have an EM data and corresponding Segmentation data. Both are arrays stored in h5 format, how can I convert them into ChunkedGraph format?
Before that, I usually used igeous to generate the corresponding mesh and skeleton, and then visualized them in the neuroglancer

[sdorkenw]

Hi Achilles, the ChunkedGraph would only be needed to facilitate proofreading. Are you trying to make your segmentation proofreadable? The ChunkedGraph is not needed if you only want to visualize your segmentation (including meshes and skeletons) in neuroglancer.

The ChunkedGraph has a separate meshing pipeline once the data is ingested. I posted some info below

For data format and conversion see here: https://github.com/seung-lab/PyChunkedGraph/blob/e9e9492da3427459484be41aa2e3dda36f8daea9/docs/segmentation_preprocessing.md
and here: https://github.com/seung-lab/PyChunkedGraph/blob/e9e9492da3427459484be41aa2e3dda36f8daea9/docs/edges.md
The ChunkedGraph indeed requires the use of Google Cloud. We have kuberenetes deployment scripts and instructions here: https://github.com/seung-lab/CAVEdeployment

[liuyx599]

Thank you for your reply. Yes, we are doing the equivalent of over-segmentation segmentation concatenation (or tracing). I now have a chunk of data image and the corresponding over-segmentation(both files stored in h5 format). We know to use neuroglancer for visualization, but we want to be able to do some merge operations of the over-segmented segments on top of the visualization (similar to Merge mode in Flywire，hold down the M key and select two segments to complete the merge operation). The above is roughly a description of my problem, is it possible to use pychunkedgraph in this case, and if so, how do I proceed?

[liuyx599]

Hi Dorkenwald, If I want to implement storing with graphene instead of precomputed, do I have to re-segmentation my image.h5? My current segmentation is equivalent to generating Affinity with U-net and then translating it into a segmentation result via a watershed approach. I may not know enough at this point, but may know that graphene may require information such as affinity to aid in generation. What approach should I take that can be carried out