Limitation of face registration and memory issues

[7thstorm]

First of all, this is a great script. Thank you.

I have encountered some issues. I'll try to explain:
below is the screenshot of nvidia-smi when I start deepstack

Now, I used the script to register 25 faces. Then I looked at nvidia-smi again

as you can see memory usage shot up and stayed there. Script took around 5 seconds to execute

Next, I ran the same registration script and did another nvidia-smi

as you can see the memory usage remained high.

To test the theory, i stopped and restarted docker container, and then instead of this script, I used the curl command to register 31 images. It was almost instant, and the memory usage remained low

Not sure, if there's some setting on my end or not

[radical-squared]

I noticed in your curl command that you're only submitting 2 images as all of your uploads are either image1 or image2. Try increasing the index (i.e. image3, image4) and running curl again.

[7thstorm]

Yes, you're correct. That was an oversight on my part.
After re-running the correct curl line, my nvidia-smi now shows 3.5GB
and it stays there... is there a way to reset the memory usage, so it does not hog the system? Because if I try to rerun it yet again, it throws an error. I have to restart container again and then it will work.. In other words, I can only run that curl once before having to restart container

[7thstorm]

Is this a memory bug in deepstack? should I open a bug over there?

[7thstorm]

closing issue here
and opening issue on Deepstack - johnolafenwa/DeepStack#93