search

neurodata / ndreg-old

NeuroData's registration library, built in python using ITK
Apache License 2.0
2 stars 2 forks source link

Compute peak RAM consumption for registration as a function of size #7

Closed jovo closed 8 years ago

jovo commented 8 years ago

perhaps we can register 2 clarity brains together at multiple resolutions, and profile the code, to get out time, peak RAM, and maybe some other stuff?

@kkutten1 do you know how to do this? if not, we can ask the other guys.

kkutten1 commented 8 years ago

We can the other guys.

kkutten1 commented 8 years ago
Resolution (um) Size (voxels) Memory (GB) Time (hour:min:sec)
25 576x480x275 26.342 03:25:08
50 288x240x138 3.326 00:41:35
100 144x120x69 0.475 00:10:18
250 58x48x28 0.048 00:02:52
jovo commented 8 years ago

that's amazing!!! now to 10 micron? seems like it might take ~625 GB of RAM and take 3 days? maybe ask da when a 3 day window exists? or, ask anthony if there is a 1 TB RAM machine that he has access to?

On Tue, Dec 22, 2015 at 9:28 AM, kkutten1 notifications@github.com wrote:

Resolution (um) Size (voxels) Memory (GB) Time (hour:min:sec) 25 576x480x275 26.342 03:25:08 50 288x240x138 3.326 00:41:35 100 144x120x69 0.475 00:10:18 250 58x48x28 0.048 00:02:52

— Reply to this email directly or view it on GitHub https://github.com/openconnectome/ndreg/issues/7#issuecomment-166629074.

the glass is all full: half water, half air. neurodata.io

jovo commented 8 years ago

@kkutten1 why don't you email the stanford guys the table you made, and see what they think?

jovo commented 8 years ago

also, did you time affine & rigid as well? i presume the above is for mask, rather than image?

kkutten1 commented 8 years ago

I probably should do the 10 micron first. Few people should be using awesomer over the holiday weekend. I'll ask Da (@icoming) about it

On 12/23/2015 12:56 AM, joshua vogelstein wrote:

@kkutten1 https://github.com/kkutten1 why don't you email the stanford guys the table you made, and see what they think?

— Reply to this email directly or view it on GitHub https://github.com/openconnectome/ndreg/issues/7#issuecomment-166810359.

kkutten1 commented 8 years ago

Yes I used the mask pipeline. The above times are for the entire pipeline (including affine and rigid). I don't have times for the affine/rigid but these steps take less than 10 minutes.

jovo commented 8 years ago

ok, "the entire pipeline" means that when you do LDDMM, you initialize with affine, which in turn you have initialized with rigid, yah?

On Wed, Dec 23, 2015 at 5:59 PM, kkutten1 notifications@github.com wrote:

Yes I used the mask pipeline. The above times are for the entire pipeline (including affine and rigid). I don't have times for the affine/rigid but these steps take less than 10 minutes.

— Reply to this email directly or view it on GitHub https://github.com/openconnectome/ndreg/issues/7#issuecomment-167005629.

the glass is all full: half water, half air. neurodata.io

kkutten1 commented 8 years ago

Yes, the entire pipeline includes

  1. Generating mask for subject
  2. Rigid
  3. Affine
  4. LDDMM
jovo commented 8 years ago

you should be sleeping :)

On Thu, Dec 24, 2015 at 5:44 AM, kkutten1 notifications@github.com wrote:

Yes, the entire pipeline includes

  1. Generating mask for subject
  2. Rigid
  3. Affine
  4. LDDMM

— Reply to this email directly or view it on GitHub https://github.com/openconnectome/ndreg/issues/7#issuecomment-167088634.

the glass is all full: half water, half air. neurodata.io

kkutten1 commented 8 years ago

Haha so should you..,

kkutten1 commented 8 years ago

Closing since done