How to choose default TEASER algorithm parameters for processing

[liuyx599]

In the README file on the homepage, kimimaro provides an example of using the Python interface with a scale of 4 and a const of 500. However, since I'm not very familiar with the TEASER algorithm, I looked at your guide Intuition-for-Setting-Parameters-const-and-scale, which states that "Most of the time, the defaults are fine and you shouldn't need to touch them, with the exception of scale and const which control the detail capture sensitivity of the procedure", meaning that in most cases, the default parameters should be sufficient.

However, in the source code

kimimaro/kimimaro/intake.py

Line 58 in ffad4a6

all_labels, teasar_params=DEFAULT_TEASAR_PARAMS, anisotropy=(1,1,1),

, the DEFAULT_TEASAR_PARAMS specifies a scale of 1.5 and a const of 300. Meanwhile, in the Igneous source code https://github.com/seung-lab/igneous/blob/76a43029642fb5ad9f72b4a89bceecdaa8bb1920/igneous/task_creation/skeleton.py#L44, the scale is set to 10 and the const is set to 10.

It seems that the default parameters in these different locations are not very consistent. As a beginner (like me, lacking some biological background), how should one use these parameters? For example, if I want to process the SNEMI3D dataset, which is 1024x1024x100 (6nm x 6nm x 30nm), what TEASAR parameters should I use?

[william-silversmith]

Oof thanks for pointing out that inconsistency. Any of those choices will be fine. Igneous parameters will extract backbone without finer details. It will cause most shorter branches to be missed. scale 1.5 will be better for extracting small branches scale 4 is intermediate and will emphasize backbone and take most branches. const is in nm, so look at the calibre of the neurites you want to extract in voxels and multiply by the resolution to get an idea of the scaling. It should be a fraction of this size. I'll try to harmonize this with the one kimimaro uses.

…

On Sat, Jul 1, 2023, 9:30 AM s12312 ***@***.***> wrote: In the README file on the homepage, <https://github.com/seung-lab/igneous> kimimaro provides an example of using the Python interface with a scale of 4 and a const of 500. However, since I'm not very familiar with the TEASER algorithm, I looked at your guide Intuition-for-Setting-Parameters-const-and-scale <https://github.com/seung-lab/kimimaro/wiki/Intuition-for-Setting-Parameters-const-and-scale>, which states that "Most of the time, the defaults are fine and you shouldn't need to touch them, with the exception of scale and const which control the detail capture sensitivity of the procedure", meaning that in most cases, the default parameters should be sufficient. However, in the source code https://github.com/seung-lab/kimimaro/blob/ffad4a641a8c668cadb6f3ec59b80b26d7f46ca2/kimimaro/intake.py#L58, the DEFAULT_TEASAR_PARAMS specifies a scale of 1.5 and a const of 300. Meanwhile, in the Igneous source code https://github.com/seung-lab/igneous/blob/76a43029642fb5ad9f72b4a89bceecdaa8bb1920/igneous/task_creation/skeleton.py#L44, the scale is set to 10 and the const is set to 10. It seems that the default parameters in these different locations are not very consistent. As a beginner (like me, lacking some biological background), how should one use these parameters? For example, if I want to process the SNEMI3D dataset, which is 1024x1024x100 (6nm x 6nm x 30nm), what TEASAR parameters should I use? — Reply to this email directly, view it on GitHub <#80>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AATGQSIVSNUHMFQABCX6WUDXOAQ6NANCNFSM6AAAAAAZ2ZWSJU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>