Crashing SSM on spectrogram

[alexarje]

I am trying to create an SSM with the spectrogram mode based on a 10-minute-long video file (the same file I have used to create SSM with motiograms). It says that it has created the spectrogram in a couple of seconds:

but then suddenly the kernel crashes without any other notice:

What is wrong?

[alexarje]

I don't get the crash if I load a 2-minute-long version of the same video (generated with a skip=5). Then it generates a SSM in a few seconds:

Is it a memory issue? Strange, though, given that my laptop is quite powerful.

[joachimpoutaraud]

It might be related to a memory issue yes, because it is necessary to compute the dot product of the arrays in order to get the SSM. Consequently, it can be problematic when using long video file. However, it should not crash for a 10-minute-long video.

I need to find a way to compute the dot product in separate thread or by splitting the arrays into multiple sub-arrays.

[alexarje]

Great. I consistently get crashes when attempting to run 10-minute videos. This is on a machine with 16GB RAM.

[joachimpoutaraud]

I have updated the _ssm.py script with a slow_dot function which computes the dot product in an iterative way in order to avoid any memory error issues. This is implemented when video files are longer than ~5minutes. The drawback is that it is much slower but it has fix this issue.