I was lurking on this forum for a few years and it helped me so much during my PhD. Today I can’t find the answer so I decided to sign up and contribute!
My goal is to perform real time 3D segmentation on light sheet fluorescence microscopy data. The segmentation task is very easy (adaptive thresholding mostly do the trick). Up to now I used iLastik in my pipeline for the segmentation in headless mode from Python.
I did an extensive search of current approaches for 3D segmentation and to my surprise, papers only compare segmentation performances but never the scalability (can the approach process large stacks) and the complexity/speed (what is the throughput in voxel/s). Ideally I would like to segment 33.5 MVoxel/s (even more because I need to process the data afterwards in real time too).
iLastik is currently too slow for my application and before trying to hardcode the random forest on GPU I am curious if you know of any papers that compare the computational performance of segmentation algorithms or if you have an idea of the best candidate for segmenting data as fast as possible.
My stacks would be 2048 x 2048 x 5000 but I can always divide them if needed.
Thanks for your help!