cannot use GPU for blob tracking

[t00f4r]

Describe the bug
GPU is working but Blob tracking won't let me set Use GPU to "TRUE".

To Reproduce
Steps to reproduce the behavior:

1. Go to 'Blob tracking'
2. Click on 'Set background videos'
3. 'Use GPU' is greyed out

Expected behavior
'Use GPU' is a drop down letting the user choose TRUE or FALSE

Screenshots

Desktop (please complete the following information):

• OS: Windows 11
• Python Version 3.13
• Are you using anaconda? Yes

[sronilsson]

Cheers @t00f4r - thanks for letting me know - just one question before I dig: which version of SimBA do you see when running pip show simba-uw-tf-dev in the SimBA conda environment ?

[t00f4r]

Hi! The version is 4.3.9. Let me know if there is anything else you need! Thanks!

[sronilsson]

Thanks @t00f4r I found the culprit, it's hardcoded disabled in the code 👍🏻

I'm not entire sure why it would be disabled, maybe there is some problem with it... let me test it tomorrow, getting late, before I reactivate it

[t00f4r]

Ohhh that's why. Alrighty. Thank you so much @sronilsson !

[sronilsson]

Hey @t00f4r,

I had made some little notes in the code for myself luckily.

For background subtraction, I have written a note for the GPU function HERE - Needs work, CPU-multicore appears faster., Which is probably why I greyed out the dropdown. I had done some calculations saying that to remove the background for about 28k smaller sized images on the GPU it takes like 3min which does seem on the slow side, unfortunately I don't seem to have noted down the runtimes on the CPU for comparison but it may be faster or very similar:

I don't know how the runtimes look for you, if very slow/unaccaptable speeds. If so, and you had a single video to share, it would be helpful for me to test on and see how I can improve the processing time on the GPU.

[t00f4r]

Hi. I have not timed my runs yet on CPU, but I can do a run, time it and send you the videos to test. You need the background and the subject video, correct?

[sronilsson]

Yea that would be helpful. Just one example. I have some test videos myself, which I found online and used to build all this code. But I don't know how well they generalize. Any videos would help me a lot to make sure the code scales and works for other people.

[t00f4r]

Hi! Sorry this took a while. One video (6min, 10827 frames) took 376.7171 s (around 6:16 min) to run background subtraction and 1430.9502 s (around 24 min) on blob tracking. Sadly the video is too big to upload here (132MB). I want to downsample it but don't know what resolution and framerate to downsample to. Do you have any suggestions?

also, I'm getting an error during visualizing the blob tracking where the .csv data file has one row less than the video. I did not tamper with the data file at all. What can I do about this?

[sronilsson]

Thanks @t00f4r - let's see if I can replicate the missing frame issue. Do you have a specific error message or warning message you see in the terminal you can share with me to see what's happening?

I am not sure what the resolution and fps is currently. I would say fps of 15 but you may be able to do with 5 depending on the use case, but it may be 5 already so I don't know until I know what your working with.

You could share the video on gdrive with sronilsson@gmail.com?