Raspberry Pi LESSON 65: Increasing Speed of Tensorflow Lite for Object Detection
ฝัง
- เผยแพร่เมื่อ 3 พ.ย. 2024
- Announcing the Most Awesome Raspberry Pi Lessons of All Times! This time we RUMBLE!
In this class series, we will be using the most excellent Sunfounder Ultimate Raspberry Pi kit, available here: (Affiliate Link)
amzn.to/3tSk9Mo
or for our UK friends, amzn.to/3I5d401
In today's video I will show you how to increase the speed of tensorflow lite running on the Raspberry Pi 4. I will show how we can tweak system parameters to increase speed from 2-3 frames per second to 7 FPS. We do this through thermal management and memory management.
If you want to grab those cool little straight jumper wires I am using to keep my breadboard builds neat and clean you can snag a box of them here:
amzn.to/36NyfHq
If you guys are interested in the oscilloscope I am using, you can pick one up here (affiliate link):
amzn.to/3v13Mhl
You guys get your hardware ordered so you can follow along at home!
You will also need a Raspberry Pi. I suggest the Raspberry Pi 4. If you do not already have one, this is the most suitable gear I could find:
amzn.to/3pBMfKm
The Raspberry Pi's are sort of pricy right now, so you can look on ebay or elsewhere to see if there are any deals. You will need a SD card. If you do not already have one, this is a good one:
amzn.to/3KcXTn2
I like using a wireless keyboard and mouse to have fewer wires. You can certainly use your USB keyboard and mouse, but if you want a nice wireless one, this one works on the pi. We demonstrate this by using a button switch to control a LED.
amzn.to/36Rv9Sk
You guys can help me out over at Patreon, and that will help me keep my gear updated, and help me keep this quality content coming:
/ paulmcwhorter
[Disclosure of Material Connection: I am a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for sites to earn advertising fees by advertising and linking to amazon.com. ]
#tensorflow
#objectdetection
#raspberrypi
![🔴 ฟุตบอลแชมป์กีฬา 7HD แชมเปียน คัพ 2024 [รอบรองชนะเลิศ]](http://i.ytimg.com/vi/zPm8ClLVhoE/mqdefault.jpg)
Great Lesson, as always! Thanks Paul. I agree, RPi Pico and Jetson Nano is the ideal combination. I would be very interested in more lessons on the Jetson Nano from you!
Thanks Paul. This was a great camera series.
Hello Sir, following you from Ghana, I last thing before you move to the hardware part; custom object detection. How to train the model on your own dataset to detect custom objects. Thanks
I love the knowledge you have! Anyway we can add a mp3 file when a object is detected so it can play the file. So like let’s say we have a “ security camera” than if it detects a person it will play hey stop or some type of warning,
I tried importing vlc player and the mp3 file and got that to work but I’m trying to get that that portion to align when the object is detected it’s seems a bit tricky so far!
Hi Paul,
I really liked your videos. Keep up all the good work.
i love your work, just discovered your channel, greetings from Romania !
Welcome!
Love your work sir, Greetings from India
When I started the app my temperature was 57 deg and I saw 5 fps. After running for a while my temp is 83 and fps is 4, so I may be seeing some throttling. I have installed heat sinks but no fan. My Pi 4 has 8GB of memory. Thanks for this great lesson. Faster is better!
would it be practical to use to notify me when a UPS truck goes by my house or would it require an expensive camera?
probably doable but you would have to get a clear view of the street which might mean puttinig the project outside, which means power and case become the question.
great lesson, on next lesson could you give us how to train the models for tflite
Are you going to give us the answer to the home work on Lesson 64 moving the Pan/Tilt camera to the detected object?
Hi Paul, I'm still struglling with this. With TensorFlowLite i have 5 fps but without at least 30. What can i do more?
You will probably need to move to a Jetson Nano.
Thanks Paul!
Waiting for this lesson
Hi Paul. Can we see a session where we control a GPIO pin on tensorflow. Detect an object and switch on an LED.
My Pan and Tilt hat does NOT have any GPIO pins mounted. I have only holes in the PCB
Hello sir i am from India 🧡🧡🧡
Perfect!
Sir teach more AI using raspberry pi like custom object detection in raspi
The "idle" temperature on my Pi 4 is 55 °C. When running the object detection script, it increases to 88 °C.
Did you do the updates I mention in this video? What FPS are you getting?
@@paulmcwhorter My EEPROM Firmware was already the latest version. I have increased the memory for the GPU. I can get up to 5.8 fps, then red lines are shown in the temperature monitor.
As always it was a great lesson! Had some trouble with the OpenCV lesson about using Trackbars, but I kept going and TensorFlow works without any problems. Here is the Homework: th-cam.com/video/qVL86xfwH58/w-d-xo.html
I´m using a Pi4 model B with 1GB, I´ve put some heatsinks on the Pi and a 50mm cooling fan.
Bill at DroneBot Workshop has released a video showing how to train a model own objects, and the use an ESP32 CAM/Eye-module to do object detection;
"Simple ESP32-CAM Object Detection"
th-cam.com/video/HDRvZ_BYd08/w-d-xo.html
Is it possible to do something similar on the Raspberry Pi?
Great lesson, I am posting homework for this week. th-cam.com/video/8X9kfp77k9w/w-d-xo.html This is from my Rockpi4c+ which I have been using for the last 2 lessons and is working better than my raspberry pi 3b, it got slow 1 fps. The rockpi4c+ has 4gb memory, it runs on debian bullseye, at the moment I am using a webcam, picamera2 will not install, not sure why.
LEGEND!