Regarding compute_ang_vel.m missing perframe feature file

[Divyansh Rana]

Hi Everyone,

I was using JAABA for training a classifier using outputs from flytracker. Some context - I have fruit flies videos (4x3 chambers) and I want to make a lunging classifier. For that when I am wanting to load the output from Flytracker to JAABA I select the target as "eyrun_flies" but JAABA while computing the missing perframe features gives me an error of compute_ang_vel.m file being missing. 

Any updates on this as to how we can overcome or mitigate this issue. I saw at some webpage that we need to rename some of the perframe files but which ones to specifically target is my concern.

Regards,

Divyansh Rana

[Divyansh Rana]

If however I am using flies it does work but using eyrun_flies as target it pops up this error. I have a pretrained classifier that I want to use but due to the target mismatch that classifier isn't working here. Any idea how can this be fixed? Maybe renaming or something?

[Alice Robie]

The ang_vel perframe feature should be computed by FlyTracker directly. It sounds like you didn't select 'save JAABA folders and its perframe features sub-option when you ran the FlyTracker. 

You can either rerun the the FlyTracker with these options selected or if you are a capable coder you can check whether these features exist in the *-feat.mat file and use writeJAABA function to create this files without running FlyTracker. 

Alice Robie 

--
You received this message because you are subscribed to the Google Groups "JAABA" group.
To unsubscribe from this group and stop receiving emails from it, send an email to jaaba+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/jaaba/73392bf5-d612-4902-816a-0ed70e510c50n%40googlegroups.com.

[Divyansh Rana]

Thanks for the reply. I did use the saveJAABAfolders while computing and also saved the perframe features alongside through FlyTracker.
But when I include it in JAABA, I select the target type as eyrun_flies and it gives me the above error but if I just select flies it is able to compute other features and works. However for my experiment and trained classifier I need eyrun_flies as target type. I did use the options of savetoJAABA and save perfumes as well still I get into this error as above. I saw in one of the threads though in the JAABA-master folder we need to change names of certain perframe files to mitigate this issue, specifically which ones is my concern.

[Alice Robie]

It sounds like you already have a trained lunge classifier and you're trying to apply this to new data? The most important thing in that case will be to have the features computed in the same way that it was done for the training data of the classifier. 

Can you tell me more about where the trained lunge classifier came from? Is this something you trained, from a publication, or from the lab you're working in?  

If you load the classifier like this in matlab load('path-to-your-lunge-classifer.jab','-mat') then type x.featureLexiconName, what is returned? And if you type x.featureLexicon.perframe what is returned? 
Can you share with me this jab file, a training expdir and an expdir that you are trying to classify?

There is a incompatibility between the perframe features in the JAABA eyrun target type and what is output by FlyTracker currently. Eyrun provided a patch for the .xml file as discussed in an earlier JAABA google groups thread, but it is important to understand the state of your jab file and data before using that fix.  

Alternatively, you should be able to create a new jab project with the flies or wingedfly target type, import the movies and labels from the lunge classifier (if you have access to those directories) and recreate a fairly similar classifier that you can use with the current output of FlyTracker. 

-Alice Robie 

To view this discussion visit https://groups.google.com/d/msgid/jaaba/f145888f-b3ea-4301-a580-31baa1091591n%40googlegroups.com.

[Divyansh Rana]

Hi Alice,

I figured out what was wrong. Basically the perframes flytracker generated were like log_ang_vel.m, norm_contrast.m and so on. So I renamed them as ang_vel.m and contrast.m, axis_ratio.m and so on. It worked and opened my video and correct labeling in JAABA.  But now I am running into another problem- I have a classifier made for lunging that someone else in my lab had created. I imported that on my video and its importing alright. But when I am predicting on my video using Classifier -> predict it does work but ends up giving me error like below. Any Idea how we can mitigate this issue.

 

[Divyansh Rana]

The run returns this below. And the perframe features computed via FLytracker are attached in the screenshot here.

>> load("/Users/divyanshrana/Desktop/Maria_SLEAP/Ashirwaad_pretrained_classifiers/lunge11thjulyupdated.jab",'-mat')
>> x.featureLexiconName

ans =

    'eyrun_flies'

>> x.featureLexicon.perframe

ans =

  struct with fields:

                   ang_vel: [1×1 struct]
             ang_vel_diff1: [1×1 struct]
             ang_vel_diff2: [1×1 struct]
             angle_between: [1×1 struct]
       angle_between_diff1: [1×1 struct]
       angle_between_diff2: [1×1 struct]
                axis_ratio: [1×1 struct]
          axis_ratio_diff1: [1×1 struct]
          axis_ratio_diff2: [1×1 struct]
                  contrast: [1×1 struct]
            contrast_diff1: [1×1 struct]
            contrast_diff2: [1×1 struct]
             dist_to_other: [1×1 struct]
       dist_to_other_diff1: [1×1 struct]
       dist_to_other_diff2: [1×1 struct]
              facing_angle: [1×1 struct]
        facing_angle_diff1: [1×1 struct]
        facing_angle_diff2: [1×1 struct]
             fg_body_ratio: [1×1 struct]
       fg_body_ratio_diff1: [1×1 struct]
       fg_body_ratio_diff2: [1×1 struct]
                  leg_dist: [1×1 struct]
            leg_dist_diff1: [1×1 struct]
            leg_dist_diff2: [1×1 struct]
              max_wing_ang: [1×1 struct]
        max_wing_ang_diff1: [1×1 struct]
        max_wing_ang_diff2: [1×1 struct]
              min_wing_ang: [1×1 struct]
        min_wing_ang_diff1: [1×1 struct]
        min_wing_ang_diff2: [1×1 struct]
                       vel: [1×1 struct]
                 vel_diff1: [1×1 struct]
                 vel_diff2: [1×1 struct]
          mean_wing_length: [1×1 struct]
    mean_wing_length_diff1: [1×1 struct]
    mean_wing_length_diff2: [1×1 struct]

On Monday, March 23, 2026 at 5:09:20 PM UTC-4 aro...@gmail.com wrote:

[Alice Robie]

Ok - these outputs tell us your labmate either used a very early version of FlyTracker (no version of FlyTracker I have access to output these features) or renamed the perframe features the way you described. I am not sure how to help you figure that out. 

See: https://kristinbranson.github.io/FlyTracker/FAQ.html
Just to be very clear: FlyTracker is exporting perframe features that have been transformed: log_ means its the log of the original value and similarly norm_ is a normalization. Just changing the file names will not make the data values match the features in the classifier you are trying to use unless that's also what the person who created the classifier did.

Re your error: there seems to be an issue with the perframe features this may be due to mismatch between original perframe features and your renamed perframe features. I can't really troubleshoot this just from a screenshot. 

Alice 

To view this discussion visit https://groups.google.com/d/msgid/jaaba/50b9c8a3-bfaf-4a35-8769-57e0c5a89d01n%40googlegroups.com.