Need help simulating .tflite yolo models on my linux machine.

Ajay Bidyarthy

@thomas
Hi thomas,
Yes, so the table I'm referring to is under the benchmark section here: https://docs.modalai.com/voxl-tflite-server-0_9/#benchmarks
So the thing is when I run the voxl-looger with -s 5000 I also expect the code to run for that many samples and output some log, but as you see from my screenshot, it just outputs ... ' samples: 0'

And no worries thomas, I do appreciate your time and help here

Regards

@Ajay-Bidyarthy

Okay yeah let's back up here - why are we even trying to get a log with IMU samples in it? If you want to benchmark voxl-tflite-server, you don't need IMU samples to do so. What I had meant by all of this was: you don't have a camera attached to your VOXL if I remember correctly and so instead of attaching one, we could use a voxl-logger log which had camera data in it for replaying. This is because when you replay a voxl-logger log, it writes over all of the pipes. If we had a log which had camera data in it, you could replay it and your VOXL would "think" it had a camera attached and so then when you run voxl-tflite-server it could use the data coming from those pipes.

In this case, I probably need to provide you with a log that has camera data in it. No worries though, I have lots of logs and can upload one here. Do you have any preferences of what's in the log? I was just going to take my drone and wave it around for a few seconds but if you'd like something more we can do that. I'll also verify that this log can be used by tflite-server and what steps need to be taken.

If that seems reasonable let me know and I'll get started.

-Thomas

Ajay Bidyarthy

@thomas
Oh I understand now. So using your camera data voxl logs. We can simly use those logs(camera data) to benchmark tfliteserver and get the benchmark stats like shown here:
https://docs.modalai.com/voxl-tflite-server-0_9/#benchmarks

I would appreciate a lot if you could share them. Since it would be good to use those logs/same camera data to compare stats from diff custom tflite models

Thanks a lot

@Ajay-Bidyarthy

Yep exactly! Let me get working on a log right now for you to use and then I'll upload it here and give you instructions on how to use it.

Thomas

@Ajay-Bidyarthy

Here's a link to a .tar with a voxl-logger log that I recorded! https://drive.google.com/file/d/1c-snBLT6mmrwGpKE2WhVR-YQgumN0s6o/view?usp=sharing

You should use ADB to push this file onto VOXL and then use tar -xvf customlog.tar to unzip the file, it should unpack into /data/voxl-logger/log0009/. Then you should be able to voxl-replay -p /data/voxl-logger/log0009/. The only other important thing you need to do is edit voxl-tflite-server to predict from the tracking camera in /etc/modalai/voxl-tflite-server.conf.

Thomas

Ajay Bidyarthy

@thomas
Hey thomas,
So I have followed your steps and loaded the log0009 data and then ran voxl-replay command while running voxl-tflite-server correctly configured with tracking option. But nothing happens after that it seems.

So the output I'm expecting at the end is to be able to have something like
Model Task Avg Cpu Inference(ms) Avg Gpu Inference(ms) Avg NNAPI Inference(ms) Max Frames Per Second(fps) Input Dimensions Source
custom-tflite-model Object Detection 33.89ms 24.68ms 34.42ms 34.86750349 [1,300,300,3]

Where can I get such output after running the voxl-replay command. I think I'm missing some steps after that.

Please let me know what I'm missing here.
Thanks

@Ajay-Bidyarthy

Yeah, so my default all of our services don't print anything out when run. Do voxl-tflite-server -t for timing printouts.

This won't give you a super organized printout like what you are writing. But it will give you out data on the per-iteration callback times and the breakdown per function. You'll then need to modify /etc/modalai/voxl-tflite-server.conf in different ways to get different timing printouts in order to build up a table of benchmarks.

Thomas

Ajay Bidyarthy

@thomas
Okay, So i did run voxl-tflite-server -t alongside voxl-replay command but am still not getting anything.
For yolov5 I get no output after running voxl replay. it just outputs this:

For my custom yolov7 nd yolov8 I get these errors:

I even referred to this thread https://forum.modalai.com/topic/2997/can-anyone-recommend-a-tflite-colab-notebook-for-voxl2-training/3

and configured the main.cpp file by adding if else statement for custom model to use YOLO model instead of OBJECT_DETECTION as the default one before even running the above commands.

I'm not sure why am I still not getting any output.
Thanks

@Ajay-Bidyarthy I'll look into it today.

Thomas

@thomas @Ajay-Bidyarthy

Sadly I won't have time to look at this today and am out of office tomorrow so I can't fully debug this for you. Let's focus just on YOLOv5 right now since that's the one we have onboard by default and it isn't segfaulting. Make sure you're launching tflite-server in a terminal and then in a separate terminal running the voxl-replay command.

I apologize for this, sorry I didn't have time for it today.

Thomas

Ajay Bidyarthy

@thomas
No worries. Ok, but even for yolov5 I've been running both voxl replay and tflite server commands in separate windows/terminals but still facing the issue.

Let me know why is it so and how can we resolve this.
Thanks

@Ajay-Bidyarthy

Hey Ajay, working on it right now. Will let you know when I have an update.

Thomas

@thomas @Ajay-Bidyarthy

Okay so I've identified a problem in our timing code and am working on a fix. Here's a short-term solution for you:

Make sure /etc/modalai/voxl-tflite-server.conf has "input_pipe" set to "/run/mpa/tracking"
Get your VOXL's IP address with voxl-my-ip and enter this IP into your web browser to pull up voxl-portal in your browser.
Run voxl-tflite-server on VOXL
Run your voxl-logger log as voxl-logger -p /data/voxl-logger/log0009/ or whatever the path is saved as

Now in your web browser, from the "Cameras" dropdown click tflite. You may need to refresh the page to get this to pop up. The permalink to this page is <YOUR IP>/video.html?cam=tflite (e.g. http://192.168.0.198/video.html?cam=tflite) so you can optionally just go there. This overlay shows the ML model running and in the upper left of the frame shows a framerate. You can use this to estimate for the time being.

Am working on the fix now, will let you know when it's released.

Thomas

@thomas @Ajay-Bidyarthy

Okay, I've pushed up a change for our nightly builds to fix the timing printouts with the -t option. So starting tomorrow you can flash the latest nightly SDK build and have this fix. If you'd like it today let me know and I can push the .deb up to Google Drive so you can flash it directly. Here is the link for the nightly SDK builds.

Thomas

Ajay Bidyarthy

@thomas
Hi,
Thanks for this info. So I have tried accessing the portal by running the voxl-my-ip command but found nothing getting printed out when I run that command. I inspected to make sure my voxl-portal service is running perfectly but still cant seem to get the ip for portal.
Also, can you push the latest SDK as well or share the deb with me.
file:///home/bhavya/Downloads/WhatsApp%20Image%202024-04-09%20at%2010.46.27%20AM.jpeg