Template picking

Hi,

I really much like the result that come out of template picking. However in many cases it is unbearable slow. I chose now 10 references and I get only 1000 micrographs per day picked. I am seeing however it only uses 4 cores and I don’t have the chance to set this higher. Would it be possible to include a variable number of cores as its done for ctf find?

Additionally, at least the output says it preprocesses the templates for every individual micrograph, I guess this causes also some of the time.

Best,

David

Hi @david.haselbach

Are you sure you’re using the latest version of cryoSPARC? since v2.0.18 the template picker has been fully GPU accelerated and should take less than 1 second per micrograph.

Hi,

Yes I even have 2.1
This is the output I am getting:

Launching job on lane default target localhost …
License is valid.
Running job on remote worker node hostname localhost
Project P2 Job J20 Started
Master running v2.1.1-embo, worker running v2.1.1-embo
Running on lane default
Resources allocated:
Worker: localhost
CPU : [0, 1, 2, 3]
GPU : []
RAM : [0]
SSD : False

Importing job module for job type template_picker…
Job ready to run


Hello world from P2 J20!
[ 0 9 11 18 20 25 32 34 41 42 43 46]
template psize 1.71875
micrograph psize 1.1
micrograph shape: (4096, 4096)
old template shape: (256, 256)
new template shape: (402, 402)
micrograph num: 8833
template num: 12
bin factor: 4
Processed 5220 of 8833 micrographs in 410430.00s
flip the templates…
processing template - 0
flip the templates…
processing template - 1
flip the templates…
processing template - 2
flip the templates…
processing template - 3
flip the templates…
processing template - 4
flip the templates…
processing template - 5
flip the templates…
processing template - 6
flip the templates…
processing template - 7
flip the templates…
processing template - 8
flip the templates…
processing template - 9
flip the templates…
processing template - 10
flip the templates…
processing template - 11
non-max suppression box size: 27
process 5220 out of 8833 micrographs
micrograph 5220 remove 0
num_result: 173
coords: (173,)
flip the templates…
processing template - 0
flip the templates…
processing template - 1
flip the templates…
processing template - 2
flip the templates…
processing template - 3
flip the templates…
processing template - 4
flip the templates…
processing template - 5
flip the templates…
processing template - 6
flip the templates…
processing template - 7
flip the templates…
processing template - 8
flip the templates…
processing template - 9
flip the templates…
processing template - 10
flip the templates…
processing template - 11
non-max suppression box size: 27
process 5221 out of 8833 micrographs
micrograph 5221 remove 0

nvidia-smi also doesnt show that the Gpus are in use.

Best

David

Any chance the job you’re running is a clone of an old picking job? The CPU-only template picker is still present in the codebase but has been hidden… can you try creating a completely new Template Picker job? if you go to the “metadata” tab in the inspection window of the job (after clicking spacebar) you should see
"type": "template_picker_gpu",

oh you are absolutely right. Thank you!