Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    JensBlack
    @JensBlack
    hey there. quick question: Are the two functions to train models "train single model..." and "Train multiple models..." saving the models in different folders? It seems that one of my models is saved in models/generated_models (trained with single model), but the other two are in models/validations/model_files?
    sashakrol
    @sashakrol
    image.png
    @sgoldenlab i got around to trying the new visualization to only use some of the plots. And got an error at the merge step.
    JensBlack
    @JensBlack
    @sashakrol looks like open cv got an empty argument passed to resize(). As it is not throwing an error earlier I guess that the image was read correctly. So the error must be in the if/else clause (line 67 to 102). Looks like in the "else clause" (line 78) both arguments "largePanelWidth" and "largePanelHeight" are not defined. They are only mentioned in the if statement above (line 72) (which are exclusive). You could try copying the line 72 and put it in the else clause
    JensBlack
    @JensBlack
    okay, now that i had a closer look at the code... (especially at the for loop above) i have to retract what i said. it would only go into that case if it already passed once through the if statement that defines both "largePanelHeight" and "largePanelWidth" (see the "largePanelFlag" setting).
    sashakrol
    @sashakrol
    @JensBlack thanks for taking a look! i had a look at the code briefly and couldn't see anything easily fixable so put a pin in it.
    JensBlack
    @JensBlack
    @sashakrol no problem. sorry, that it didn't help though
    i had a lot of issues with openCV for deeplabstream, so i thought it would be worth a try :D
    sgoldenlab
    @sgoldenlab
    @sashakrol @JensBlack - thanks for letting me know I'll take a look now !
    sgoldenlab
    @sgoldenlab
    image.png
    It looks like the code can't find yous sklearn plots. While I'm looking to recreate the error: I did make a change to the folder names that could be relevant. The folder created when generating the frames (inside of the project_folder/frames directory) previously had a name added to them to indicate which kind of plots they were, e.g., "Video_1_gantt", "Video_1_path", "Video_1_sklearn". This made it trickier to find them so I removed it. Instead the frame folders are called just "Video_1"
    In your "project_folder/frames/sklearn_results" folder, could you rename the folder to just the video name and try again?
    sashakrol
    @sashakrol
    unfortunately the folders were already named video_1 within sklearn_results, probability_plots and gantt_plots
    sgoldenlab
    @sgoldenlab
    Just to check: inside the folder "Video_1", inside of sklearn_results, you have image png files like this starting with "0.png"?
    sashakrol
    @sashakrol
    yes
    sgoldenlab
    @sgoldenlab
    Does this happen on the very first image? ie you don't see any images being created like this?
    image.png
    sashakrol
    @sashakrol
    yes - no images being created.
    a merged folder is created ( though not surprising/helpful I guess since that's early on. )
    sgoldenlab
    @sgoldenlab
    In merge_frame_movie.py - could you help me bt inserting two pint statements like this (the row numbers may be changed ever so slightly from the version you have), and tell me what paths are being printed out in the SimBA terminal window?
    image.png
    It seems like the image that is being joined to the main sklearn prediction frame does not exist or can't be found
    sashakrol
    @sashakrol
    image.png
    oh!
    the probability_plots is in a subfolder within the folder
    within the folder entitled Video_1
    called the thing the classifier is called
    let me move the images ( will take a bit of time) sounds like that should fix it.
    sgoldenlab
    @sgoldenlab
    ah damn I might had changed that too sorry - can you remove that subfolder (named after the classifier) and place the folder named after the video directly in the probability plots folder
    yes
    sashakrol
    @sashakrol
    Thanks for the troubleshooting!
    (for what its worth: minor aesthetic feedback: the probability line plots are rather thick looking. Not as pretty as in the original validation plot) =)
    sgoldenlab
    @sgoldenlab
    Thanks! I'll reduce it. The linewidth is set in plot_threshold.py, on line 34:
    image.png
    sashakrol
    @sashakrol
    I think its working!
    sgoldenlab
    @sgoldenlab
    fingers crossed!
    sashakrol
    @sashakrol
    yep. it was that extra classifier folder.
    sgoldenlab
    @sgoldenlab
    :thumbsup:
    meghanflanigan
    @meghanflanigan
    @sgoldenlab , it seems that simba is getting stuck on the analyzing machine predictions step. The distances/velocities were analyzed just fine, but when i select the machine prediction parameters I would like analyzed and press analyze, it just sits at video 1/9 and doesn't seem to do anything (these steps were fairly fast when I ran them before so I am assuming it's stuck if nothing has happened for ~30 mins). Any ideas what this could be?
    image.png
    (no errors showing in the terminal here)
    sgoldenlab
    @sgoldenlab
    @meghanflanigan - let me check on this, as you say it should not be this slow, although it's significnatly slower than the movement/velocities. How long are your videos, and how many classifiers are running on each video?
    sgoldenlab
    @sgoldenlab
    @meghanflanigan - I have a method for significatly speeding up this part of the code, I will upload it and let you know
    sgoldenlab
    @sgoldenlab
    @meghanflanigan - I have updated the code to be "analyze machine results" script to be more efficient - would you mind downloading SimBA again, load your project, and see if it is quicker?
    meghanflanigan
    @meghanflanigan
    yes i will try this now, thank you! I let it sit for at least an hour and it never moved past video 1. My videos are 800x1100, 30 fps, 10 minutes
    5 classifiers/video
    meghanflanigan
    @meghanflanigan
    @sgoldenlab it worked right away! thank you!
    meghanflanigan
    @meghanflanigan
    The values that came out of this don't appear to match was was calculated in the validation videos, and there are many negative numbers and things that don't make sense (for example, total duration being ~1.5s but the mean bout duration being 7.5s, # of events being decimals). Could there have been some error in making the table?
    sgoldenlab
    @sgoldenlab
    @meghanflanigan: Yes it could possibly be a mistake by me in creating the table. I will look over this now!
    meghanflanigan
    @meghanflanigan
    great thanks! And just to clarify, in the movement log, the last columns where it says mean/median animal distance, this is the distance between the two body parts that I specified when I ran the movement analysis?