Best way to do surface distance measurement

[linhua-tai]

Hi Mart,

I have couple of questions on what's the best way to get void segmentation.
My tomogram is at 10 A/px but the header after isonet denoising is 1 A/px. I used this command
'easymode segment void --data isonet-misalign --2d --tta 8 --output misalign-segmented --apix 10 --gpu 0,1,2', the output is:

GPU 0 - starting inference with 2D model 'void' at 20.0 A/px (n_workers = 16, tta = 8).
1/1 (GPU 0) - void.scnm - hela-wt-20251025-L10_00_ts_060_10.00Apx.mrc - eta 00:00:00 (43:37 this tomo)
void: using 2D model
easymode segment
model_path: void
input_data: ['test-void-segment']
output_directory: test-void-rightpixel
gpus: 0
tta: 8
overwrite: False
ais_2d_nets: True

Found 1 tomograms to segment.

ais segment -m /home/tai/easymode/void.scnm -apix 20.0 -d test-void-segment -ou test-void-rightpixel -tta 8 -p 1 --overwrite 0 -gpu 0 --stride 1 --use-depth 1.0

It appears to me that easymode called ais and ais -apix 20.0 confused me, since I have --apix 10 for easymode. I was wondering if it's ok to run void segmentation on 10 A/px tomograms, and should I change the header to 10 A/px?

Best,
Linhua

[mgflast]

Hi Linhua,

I'd say the best way to go about something like this is to write the correct value for the pixel size in the metadata. For segmentation we rescale the tomograms to whatever pixel size a model was trained on (mostly 10 A/px, but some models operate at 20, 30, or 50; void is at 20). In the code we assume that the header pixel size is correct.
When it isn't, you can indeed add the --apix 10.0 flag to tell easymode what the actual pixel size is. It looks like in case of a 2/2.5D model (for which we use some Ais code) that argument doesn't get forwarded properly. I'll fix that but in general and for now, best just set the pixel size correctly to avoid this sort of thing; I don't want to change things in easymode just because there's a bug in IsoNet2 🙃

[linhua-tai]

Hi Mart,

thank you for the explanation! I'll give it a try on 20 A/px tomograms then. Another thing, is there a way to check the quality of lamella mask after void segmentation? It looks to me that this command:pom contextualize --starfile xx.star --samplers void:0.5:-2 .... just produces star files?

Thank you very much!

Best,
Linhua

[mgflast]

Hey Linhua, Two things: 1. about the pixel size: regardless of what apix the networks were trained on, you can run inference on tomograms reconstructed at any arbitrary pixel size, as long as the pixel size in the tomogram's header is correct. So they don't have to be 20 A/px tomograms. If you would run 'easymode segment void' on a 20 A/px tomogram that still has 1 A/px in the header, that still wouldn't work. Just as long as the header pixel size is correct you don't really need to think about pixel sizes. 2. I personally really like using Pom to inspect tomograms and segmentation outputs side by side. See the tutorial here: https://mgflast.github.io/easymode/user_guide/examples/easymode_and_pom/#easymode-pom - it shouldn't take too long to set up (once you get the hang of it, setting up the app takes just a few minutes). For a more thorough inspection of the segmentation output I tend to use something like 3dmod or ChimeraX. In general I don't recommend blindly trusting the segmentation output (or the output of any other automated tool) - always inspect a couple of volumes to check if things make sense. Or you can inspect the starfile itself. You can open it with ArtiaX and colour the particles by their distance to the lamella surface. If you do this for a couple tomograms that contain a lot of ribosomes, the pattern should be pretty clear. See panel G in this image for an example - that's a pom-contextualized ribosome star file displayed in ArtiaX, coloured by the distance to surface. [image: image.png] good luck! :) Op vr 5 jun 2026 om 15:28 schreef linhua-tai ***@***.***>:

…

*linhua-tai* left a comment (mgflast/easymode#18) <#18 (comment)> Hi Mart, thank you for the explanation! I'll give it a try on 20 A/px tomograms then. Another thing, is there a way to check the quality of lamella mask after void segmentation? It looks to me that this command:pom contextualize --starfile xx.star --samplers void:0.5:-2 .... just produces star files? Thank you very much! Best, Linhua — Reply to this email directly, view it on GitHub <#18?email_source=notifications&email_token=BSQIFOGXU7JF6LXXXELZCE346LKHJA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRTGI3DCOJWGE32M4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4632619617>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BSQIFOCUIH63LMUVWMVYHHT46LKHJAVCNFSM6AAAAACZ27D7RCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DMMZSGYYTSNRRG4> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/BSQIFOEL5MHTQJVZS5LSE7L46LKHJA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRTGI3DCOJWGE32M4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJKTGN5XXIZLSL5UW64Y> and Android <https://github.com/notifications/mobile/android/BSQIFOCTBHCI55N34KDGIZ346LKHJA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRTGI3DCOJWGE32M4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJLTGN5XXIZLSL5QW4ZDSN5UWI>. Download it today! You are receiving this because you commented.Message ID: ***@***.***>