Template matching on GPU & Dynamo

[JuhaHuiskonen]

Is it possible to run tomoBEAR template matching (GPU), and write the results to Dynamo catalogue? My project is now organised in a Dynamo catalogue, but the template matching (CPU) is too slow so I am looking for alternatives for this step. Thanks!

[MishaKudryashev]

Hi Juha,
Yes, there are some ways to run this GPU-template matching, the simple one is a function in a for-loop. If you activate tomoBEAR and deactivate dynamo it would be something like this

for i = 1:tomo_number
gpuDevice(1);% allocate a device, numerates from 1 (GPU0 -> device 1 and so on)
pts = dynamo_match(tomo,template,'mask',mask,'outputFolder',['TM_gpu/tomo_' num2strtotal(i,3)],'ytilt',maxangles(i,:),'sc',[1160 1200 1000],'cr',360,'cs',10, 'ir', 90, 'is', 10, 'bin',1);
gpuDevice([]); %release GPU
end

It will produce you an output like dynamo_match on CPU - a variable that can be treated by dynamo particle extractor and the folders with the CC-volumes and angle-volumes.

Its a good idea to be able to import a dynamo Catalogue for GPU-based template matching and table generation, particle extraction, creation of the first project, etc. We will try to create a point of entry at this step. It will take some testing, we will write it here.

Hope its useful,
MIsha

[JuhaHuiskonen]

Hi Misha,

Thanks, that seems pretty straightforward! This is my current script for running dynamo_match. It's not fully tested, but it seems to be doing what I want: keep only views close to side views (60 <= tilt <=120), write a table, and then import the table to the current catalogue.

oc = dcm('c', 'warp_catalogue','l','t')
for i = 1:200
fprintf('Starting template matching using tomogram %d', i);
pts = dynamo_match(oc.results{i}, 'template.mrc', 'mask', 'mask.mrc', 'outputFolder', 'particles', 'ytilt', [-60,60], 'sc', [400,400,250], 'cr', 360, 'cs', 15, 'bin', 1, 'mw', 16);
myTable=pts.peaks.computeTable('mcc',0.3);
idx = myTable(:,8) >=60 & myTable(:,8) <= 120;
newTable = myTable(idx,:);
peaks_file = sprintf('tables/peaks_%d.tbl', i)
model_name = sprintf('ccpeaks_%d', i)
dwrite(newTable, peaks_file);
if height(newTable) > 0
m=dynamo_model_import('c', 'warp_catalogue', 'i', i, 'table', peaks_file, 'mn', model_name);
end
end

Best wishes,
Juha

[MishaKudryashev]

Hi!
Yes, I think you need to skip 'mw', 16 - its paralellzation in Dynamo. And you need to keep an eye that when you run dynamo_match you run it from tomoBear, for the other functions you may need to re-activate dynamo.

Uh, its not pretty, we will think how to run it nicer.

try it, hope it works!
Misha