Trainable weka segmentation slower on Mac OSX

Hi, I’ve recently been using the excellent TWS plugin for biological image segmentation. I have a macbook pro which is a few years old but decent spec for the time, with 16gb ram and a 2.6 GHz Intel Core i7 processor - but also have access to a newer Mac Pro with 64gb ram etc. When I use either of these machines to run pretty simple TWS classifiers (eg only using gaussian blur and sobel filter) on multiple image files within a folder - the process is very slow, or sometimes won’t run at all.

I have windows 8 intalled on my macbook pro with bootcamp, and last night tried running a classifier (which I’d just tried to run unsuccessfuly on Mac OSX) in the Windows version of FIJI. It managed to churn through over 600 1000x1000 pixel images in about 30 minutes without any issues (other than inexplicably not being able to apply the classifier to a small number of the files on the first run).

I’m guessing most of you will be thinking “why don’t you just use Windows?!” (and perhaps why do you even have a mac in the first place…!) - and it’s likely that I will from now on, but I’d be interested to know if anyone has any experience of using TWS on the mac OSX version of FIJI - and if so, should it be possible for it to run as quickly and smoothly as it does on Windows. Clearly it doesn’t appear to be a hardware issue - could it be an issue with JAVA version for example? I think I’m currently running with the legacy JAVA version 6 on the mac, which seems to run things relatively stably, but probably not perfectly.


1 Like

A performance difference of that magnitude between OS X and Windows is certainly unexpected. There are many ImageJ users running OS X, including many in my group who use the TWS plugin.

You could try taking a thread dump to see what operation is taking ImageJ so long.

Incidentally: if you click the TWS window’s “Settings” button, are only the five default training features checked? If your OS X machines have different features checked than your Windows machine, that would explain the performance difference.

1 Like