Applying a classifier to an image

[davodogster]

I specified 36GB RAM max when applying a classifier, however CATS split the image into too many tiles and only used 7.5GB RAM at the most during this time, resulting in a very slow classification of my 3GB stack (which consisted of 3 channels for one image).

[tischi]

This can be! This is tricky business, because when I am trying to be too close to the total memory it could crash. That's why I have to be conservative ( I think ) ....
I have to admit that CATS in general is quite slow. That is why we typically run it on a (linux slurm) computer cluster.

But, can you please post here the text from the IJ log window when you are running above tasks? Maybe I have some ideas...

[davodogster]

It's great that CATS automatically tiles for classification but I think you can get away with splitting into less tiles (if you think it will speed up classification). I just run another classification using "apply classifier" on a 3800 x 4800 4 channel image. It split it into 16 tiles, used 3 threads per tile, appears to have computed the "310 active features" quickly, but spend 7.5 mins classifying the image and the max RAM it used was only 8GB/50GB.

[tischi]

Could you please post here the text from the IJ Log window (just copy and past the whole thing). I need to see this to better understand what's going on.