YOLOv5 object detection on Raspberry pi 4

[@ameer-alahmadi]

Hi, first thank a lot fore your explanation. I just want to know for the raspberry pi does it make a difference in performance if I use 2, 4, or 8 GB RAM?

[@Waja_Dadkarim]

Thanks, dude it's working. 5x slower than the MobileNet algorithm that I used. Needs some improvements. Raspberrypi 4B 2GB RAM Pi camera armv7l

[@readystudygo5091]

nice video BTW can we also use a raspberry pi 3 model B+ for the custom object detection ? and can you also record a video for this as well by using the raspberri pi? Thanks

[@narpat007]

Excellent ! I have just three questions; 1. can it handle realtime 1080p h264 video stream or atleast 720 h264 video stream ? 2. What is the maximum frame rate achievable with Pi4B 4GB as it is ? 3. what is the cheapest solution to get framerate atleast 25pfs ?

[@Unfreeze007]

Same Questions, did you get your answers?

[@vungochoa6882]

Cùng câu hỏi. Mình đang chạy yolo v4 trên pi 4 4gb sử dụng cam 1080p đang fps có 0,6. Bạn thực nghiệm 360 chưa chạy ok ko

[@nasimreza9492]

hi, How can we do it with custom data set? It is already a pre-trained, model. Is it possible to use our own dataset?

[@zubaidahsabah7948]

great work! how can we implement this using tiny version, i suppose it would be faster

[@himomhey3437]

How many Gb for a raspberry pi to run YOLOv5 like in this video?

[@jarbozhang3682]

thx for the video, can you use the sci camera with this 64bit version system? i have found many solutions, but no one is useful. I guess that's the reason why you use the usb web camera?

[@faceprofesor]

Thank you for the video. How do you pass the KZbin image through the yolo?

[@nanakwamekankam1281]

I get a "Segmentation Fault" error when I run the inference. Any help?

[@f_bian]

We can't install torch, it gives the following error.. Do you have any idea on how to solve this? Thanks! Could not find a version that satisfies the requirement torch>=1.0.0 (from thop) (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2)

[@SoftwaresAI]

I already have a tutorial about it here, kzbin.info/www/bejne/rZarp3xsn5mCipo pytorch installation tutorial for raspberry pi. Please subscribe the channel :)

[@Supik.]

Is it the same way for the yolov3-tiny?

[@singgihbekti969]

can we deploy in raspberry pi 3? does it support with pytorch? or any pytorch version? thank you

[@나현강-b6r]

Did you install the 2.2 version of tension flow?

[@rielt12]

did you use a movidius stick?

[@عبداللهالحريبي-ص1ن]

I am facing problem with installing opencv-python , it sys that object is not iterable , how can i fix it , if would you please

[@gbox3d]

great work!!

[@SoftwaresAI]

Thanks a lot!

[@swoffiam7561]

i got this error : No module name Ultralytics but in your video, you not install Ultralytics Module. How your code still works properly

[@denigajah6369]

late reply but just pip3 install ultralytics

[@betulsahin5419]

How can i download yolov5 to raspberry pi 3b+ I couldn't find anywhere. Can you help me about this?

[@luizcarlosmarquesjunior.9468]

Great video! What was the FPS for real time detection on the Raspberry Pi?

[@SoftwaresAI]

FPS changes with image size, and model you choose to predict. i guess using 360p resolution, it was taking 1.2 sec to process one image on yolov5s.pt model

[@vishalsinha8614]

Were you able to try peripherals to improve the FPS by any chance? Something like Coral accelerator?

[@ngocluu8366]

@@vishalsinha8614 try to overclock your rasp4, you will need a fan for the cpu and make it detect every 3 frames instead of every frame like the default way. by this, i have archived 5fps on rasp4

[@minhhuy499]

@@ngocluu8366 so what is your over_voltage and arm_freq in config file?

[@kashishnaqvi4708]

Hey amazing tutorial, can you help me out on how to provide audio output for the detected objects? which code to edit and how to edit? Thanks in advance!

[@danani3199]

Did you ever figure it out for the audio output?

[@chandanan953]

Did you figure out the solution for audio output?

[@konkoboaxel8887]

hi, I tried to install the yolov5 but at the step "python3 detect.py --source data / images --weights yolov5s.pt --conf 0.25" my ssh disconnects at fusing layers ... how to solve this please

[@SoftwaresAI]

When you connect raspberry pi using ssh, it doesn't have a display, so, it crash when you run the program. Try to connect to your pi device using remote desktop, and then run the program from remote desktop then it should work. I have a tutorial for that here: kzbin.info/www/bejne/iKezf5aVaMaVj9U

[@konkoboaxel8887]

@@SoftwaresAI I use vncviewer, but it disconncects

[@SoftwaresAI]

Is there any error message? Is raspberry pi reboots, or just vnc disconnects?

[@konkoboaxel8887]

@@SoftwaresAI I think the raspberry is restarting because everything disconnects the ssh, the remote desktop(vncviewver), without error. But I am testing xrdp and it gives me errors "...... magic_number = pickle_module.load (f, * * pickle_load_args) EOFError: Ran out of input "

[@SoftwaresAI]

@@konkoboaxel8887 Is raspberry pi connected to internet? when you run the code?

[@ajsirhzz75]

Does it run in 32 bit bulleyes os??

[@kashishnaqvi5639]

Hey in mine, it says no command pip3 pls help

[@merondesta3895]

can i do that using raspberry pi 3?

[@smitpanchal9866]

I can't install pytorch on raspberry pi 3 B+ ...!

[@kashishnaqvi4708]

I want to do this on my raspberry pi 3b and give an audio output for objects for blind people. Can you help me?

[@SoftwaresAI]

i will make a tutorial video about how to get audio output from yolov5 object detection in few days. Please support my channel at www.buymeacoffee.com/biplob004

[@kashishnaqvi4708]

@@SoftwaresAI That's great. Can you check if I'm added to your supporter list? I don't want to sound impolite however I need to learn it for my school work and present it really soon. How long would it take for the video? and can you tell me how can we add our own dataset to the existing coco classes? like if i want to add classes for currency as well?

[@SoftwaresAI]

@@kashishnaqvi4708 Hello, here is code that output audio : drive.google.com/file/d/1p2Hrs7tEOBZG2n-uQ8LuaY4lZc3fJX-w/view?usp=sharing

[@SoftwaresAI]

Just run the run.py file and it will start working if all required python libraries are installed. Else see error message and download required python module/library.

[@SoftwaresAI]

I have code update, with little more code, let me know, if you need it.

[@aysebeyzaunal3472]

are you using a virtual space to do the arguiments? like in ubuntu with virtual box

[@SoftwaresAI]

no

[@aysebeyzaunal3472]

can i do it in windows?

[@SoftwaresAI]

@@aysebeyzaunal3472 yes you can, you need to install pytorch

[@aysebeyzaunal3472]

@@SoftwaresAI ok thanks

[@aysebeyzaunal3472]

@@SoftwaresAI sir, can i use an arduino leonardo with tensorflow lite? Because in fonction of the object , I have to light the leds.

[@thanhan4046]

Can you give me link your raspberry OS you using???

[@viktorrrrrr]

Hey guys, I get a silly question. How can I know the IP of raspberry pi on my laptop?

[@kaveeshamadhubhashana3094]

bro please can you tell me how can i attach custom trained dataset inside to the this projects

[@jervx829]

train on colab, then download the weights best.pt. also download yolov5 repo. then use this to load the yolov5 repo locally and the weights torch.hub.load( YOLO_DIR, 'custom', path=WEIGHTS_DIR, source='local', device = 'cpu', force_reload=True )

[@megauti]

It is too (extremely) slow on R-pi 4.

[@asdftw6146]

i need help! i keep getting this even running on python 3.7.3 on my rpi. File "detect.py", line 73 stride, names = 64, [f'class{i}' for i in range(1000)] # assign defaults ^ SyntaxError: invalid syntax

[@이진-r4e]

I think you should check on detect.py line 73 and see if there is invalid syntax literally... cause I think there should not be a comma, there should be a ; or a newline... like "stride, names = 64; [f'class{i}' for i in range(1000)] # assign defaults"

[@ilhancakmak79]

How can i use it by webcam? Thanks sir

[@MrDUMBINATOR]

Hi, im stuck in the installation process for pandas, this happened with opencv as well. Can you help? Requirement already satisfied: six>=1.5 in /usr/lib/python3/dist-packages (from python-dateutil>=2.7.3->pandas) (1.12.0) Building wheels for collected packages: numpy Building wheel for numpy (pyproject.toml) ... -

[@nanthini2898]

Bro its asking username, password why its asking in terminal

[@ayamejri5183]

raspberry pi 4 4 gb ram or 8 gb ?

[@kashishnaqvi4708]

Can I use this on raspberri pi 3 b ?

[@TechToLife2024]

Nice Video , How to use edgeTPU USB from google product? sir

[@kotraner]

any suggestion to increase fps...? its too slow even with tiny weight

[@o__bean__o]

model quantization

[@o__bean__o]

Dynamic range quantization 4x smaller, 2x-3x speedup CPU Full integer quantization 4x smaller, 3x+ speedup CPU, Edge TPU, Microcontrollers Float16 quantization 2x smaller, GPU acceleration CPU, GPU

[@장병희-y2p]

Everything's going well, but when you predict it with a webcam, Raspberry Pie doesn't show the video in real time and just finishes detecting it. Is there a solution?

[@SoftwaresAI]

instead of doing --source data/images do it like --source 0

[@장병희-y2p]

@@SoftwaresAI Thank you for your prompt reply!!! I know as you replied, but only this result comes out.

[@SoftwaresAI]

@@장병희-y2p Is it fixed the problem?

[@장병희-y2p]

@@SoftwaresAI drive.google.com/file/d/12K38FEvhnE5FnUYfKlG0aTWKNldRfmNe/view?usp=sharing

[@장병희-y2p]

@@SoftwaresAI It's the same situation as the picture.

[@shahedobaid1885]

sir why do i get "Illegal instruction" when i run the file??

[@heiko287]

probably pytorch not compatible to your raspi achitecture. Thats what happens to me. You should uninstall it and find a comptible version of pytorch, if it is still showing "illegal instruction". You should accept and use other libraries for object detection such as tensorflow or openCV, etcc

[@MahyaGholami-fb7hh]

its better to install the libraries with requirement.text first .😁 "pip install -r requirement.text"

[@mertenesyurtseven4823]

Greetings from Turkey! I am a member and mentor of my FRC team "Conquera". We are using yolov5n on rasbery pi 4 8gb with 208 img size which provide 7 fps. But we need more. Can you help?

[@SoftwaresAI]

u can use jetson nano to speed up.

[@sshubam]

main reason would be because the webcam you are using does not have h.264 encoding so its computationally heavy on the pi. You should use the official raspberry pi 4 camera which uses h.264 encoding and will probably give you a much higher frame rate.

[@sshubam]

else use tensorflow lite

[@ElectroCoderEC]

why is so slow??

[@abdurrazzak1612]

cool and greate

[@SoftwaresAI]

Thanks

[@yilberrojas8306]

Good.

[@SoftwaresAI]

Thanks

[@yilberrojas8306]

@@SoftwaresAI !pip install -U -r requirements.txt

[@huyvanba2943]

I can detect with image, video smoothly but when use the camera ( source 0), it have a error like this img drive.google.com/file/d/1mSL7vomkHu60cgqiNOU4YAiQp4r0a06U/view?usp=sharing