Anyone have done timing analysis processing the sample dataset in clusters

Examples are like performance increase for clusters for large dataset (number of images in clusters timing analysis)

At what number of images is that clustering has no effect?

1 Like

Hmm, very interesting question. The only serious performance profiling that has been published (that I know of) is in this project:

I am currently writing up some of my thoughts on processing larger datasets here:

I haven’t written much yet on my findings with split merge, but the main recommendation is: make the splits as large as you can, given your memory limitations in order to minimize artifacts in the data.

I would love to see some performance profiiling given different split sizes. That would make an excellent addition to the odm-benchmarks project.

2 Likes

Agreed! I have some ODM time coming up in September to devote to docs and benchmarks, and I will put this on my list. (Also on my list is a more public way to track and manage ideas for the benchmarks project… github issues I guess?)

2 Likes

Feature requests in github issues, I suppose. We don’t have anything more sophisticated at this point… .

1 Like

Roger that. Might as well start with the available tools. Github issues is a great fit for this, IMO. I’ve just been lazy and putting them in a local doc.

1 Like