Velocity field from GUI and SCRIPT not the same.

[Jonathan S.]

Dear Community,

on my dataset, I executed PIVlab via graphical user interface (GUI) and via matlab script. I compared the two velocity fields and found that they are close, but not quite the same. 

The GUI analysis seems more accurate. I chose each parameter I could choose in the script to match what I was entering in the GUI. You can find the settings and the 2 frames attached to this post. I was wondering why they differ, and thinking that they are more settings you could select/unselect in the GUI, which are not available in the script. I would like to have the same accuracy by just executing the script. I appreciate your help.

Best wishes,

Jonathan

[Jonathan S.]

[Jonathan S.]

Jonathan S. schrieb am Freitag, 19. September 2025 um 18:09:13 UTC+2:

[Jonathan S.]

Sorry, it was not possible to post all files in one message.

[William]

Could you post an image of both results, so we can see how different they are?

[Jonathan S.]

This is the code for the PIV analysis. I am plotting the original vectors

%preprocess images

image1_uint16 = uint16(65535 * mat2gray(image1));

image2_uint16 = uint16(65535 * mat2gray(image2));

image1_pre = preproc.PIVlab_preproc (image1_uint16,...

preprocess_setting{1,2},...

preprocess_setting{2,2},...

preprocess_setting{3,2},...

preprocess_setting{4,2},...

preprocess_setting{5,2},...

preprocess_setting{6,2},...

preprocess_setting{7,2},...

preprocess_setting{8,2},...

preprocess_setting{9,2},...

preprocess_setting{10,2});

image2_pre = preproc.PIVlab_preproc (image2_uint16,...

preprocess_setting{1,2},...

preprocess_setting{2,2},...

preprocess_setting{3,2},...

preprocess_setting{4,2},...

preprocess_setting{5,2},...

preprocess_setting{6,2},...

preprocess_setting{7,2},...

preprocess_setting{8,2},...

preprocess_setting{9,2},...

preprocess_setting{10,2});

% Perform PIV analysis using the preprocessed images

[x, y, u, v, typevec,corr_map,~] = piv.piv_FFTmulti(image1_pre,image2_pre,... % typevec tells us for the calculation stage the status of the vector. 0 valid, 1 excluded by std test, 2 by median test, 3 interpolated bc of missing data

piv_setting{1,2},...

piv_setting{2,2},...

piv_setting{3,2},...

piv_setting{4,2},...

piv_setting{5,2},...

piv_setting{6,2},...

piv_setting{7,2},...

piv_setting{8,2},...

piv_setting{9,2},...

piv_setting{10,2},...

piv_setting{11,2},...

piv_setting{12,2},...

piv_setting{13,2},0,...

piv_setting{14,2},...

piv_setting{15,2}); %actual PIV analysis

% Postprocessing

[u_filt,v_filt] = postproc.PIVlab_postproc(u,v, ...

posprocess_setting{1,2},...

posprocess_setting{2,2},...

posprocess_setting{3,2},...

posprocess_setting{4,2},...

posprocess_setting{5,2},...

posprocess_setting{6,2},...

posprocess_setting{7,2});

% typevec is not refreshed in the postprocessing stage, so we do it manually.

% Vectors filtered by postprocessing are overwritten with NaN.

typevec_filt = typevec; % initiate

typevec_filt(isnan(u_filt))=2; % overwrite all newly filtered values (now Nan) with 2, although they could also be excluded due to reason 1

typevec_filt(isnan(v_filt))=2;

% typevec_filt(typevec==0)=0; % amin wanted this for some mask... but this will be confusing, because all the vectors which have been excluded in postprocessing will now be marked as normal

% interpolate missing data

if posprocess_setting{8,2}

mask_interp = isnan(u_filt) | isnan(v_filt); % places that were NaN before inpainting

u_filt=misc.inpaint_nans(u_filt,4);

v_filt=misc.inpaint_nans(v_filt,4);

typevec_filt(mask_interp) = 3; % mark as interpolated

end

Jonathan S. schrieb am Montag, 22. September 2025 um 09:30:25 UTC+2:

[Jonathan S.]

These are the two velocity fields, automatic scaling off. For later frames with worse contrast both algorithms perform way more differently.

William schrieb am Samstag, 20. September 2025 um 15:30:54 UTC+2:

[William]

Hi, I would say that you're mainly seeing random noise. Your interrogation area is way too small, and the images have a pretty bad S2N. Subtle changes in image intensity may lead to different results here (e.g. mat2gray might have an effect, or the auto contrast stretch in the GUI).

I would recommend settings like in the attached image.

[Jonathan S.]

Thanks a lot William. I really appreciate your help. It is amazing for an open source software that you are there to help us out. PIVlab helped me a lot in my research. Have a wonderful day!

Best

Jonathan

[Jonathan S.]

I just wanted to add:
With your proposed more coarse grid the two methods look much more similar, especially for high-contrast frames in the beginning. However, the GUI still seems to perform better, especially concerning postprocessing. Some low contrast frames have extrem velocity vectors that the script-based version does not filter out very well.