Real-time YOLOv4 Object Detection on Webcam in Google Colab | Images and Video

Learn how to implement YOLOv4 Object Detection on your Webcam from within Google Colab! This tutorial uses scaled-YOLOv4, the most fast and accurate object detection system there currently is. Perform object detections in real-time on webcam images and video with high accuracy and speed. ALL WITH A FREE GPU!
#yolov4 #objectdetection #cloud
THE GOOGLE COLAB NOTEBOOK:
colab.research...
In this video I cover:
1. Setting up Colab Notebook and Enabling GPU.
2. Cloning and Building Darknet for Running YOLOv4.
3. Downloading Scaled-YOLOv4 pre-trained model file, the best object detector there is.
4. Custom Functions to run YOLOv4 with Python in Google Colab.
5. JavaScript code to access local machine's webcam for images and video.
6. Running scaled-YOLOv4 object detections on webcam images and video in real-time.
-----------------------------Resources-----------------------------
Github Code Repository (yolov4-webcam notebook): github.com/the...
Tutorial for YOLOv4 Pre-trained Model, Running on Video, Formatting Output and Detections etc.: • YOLOv4 in the CLOUD: I...
Train Your Own YOLOv4 Custom Object Detector in the Cloud: • YOLOv4 in the CLOUD: B...
Official Scaled-YOLOv4 Paper: arxiv.org/abs/...
If you enjoyed the video, toss it a like! 👍
To Subscribe: / @theaiguy
Thanks so much for watching!
- The AI Guy

[@TheAIGuy]

Let me know if you have any suggestions for future videos??? As always, comment down below if you run into any issues! Hope you enjoy :)

[@shiffin_chippe]

Yolov4 on Android

[@dieweltentdecker5878]

Can you make/know a tutorial on how to make an ai that can decide if an image is good or not? For example, I have two folders good/bad in these folders are good and bad images.

[@TheAIGuy]

@@dieweltentdecker5878 what qualifies as a good or bad image? Like the quality of the image? If it’s blurry then its “bad”? Let me know!

[@dieweltentdecker5878]

@@TheAIGuy I scraped a lot of Instagram images. Some of these I think are bad (These are in the bad folder). The other images I think are cool Instagram images (These are in the good folder). I have so many pictures on my phone. I want to automate the whole choosing process.

[@shiffin_chippe]

@@dieweltentdecker5878 Mention the factors that makes an image bad or good

[@nahinahmed8630]

please make a separate/ extended video for using custom obj detection on this same colab using webcam

[@yoshithaab7290]

Can you please help in providing the code, that does object detection. The present code helps in capturing picture, but it isn't able to do object detection. Please look into it!

[@Usefi13]

I agree I am trying so much

[@shiffin_chippe]

Don't know how, but you always save me at the right time ❤️...Please make a video on implementing yolov4 in android...Love you alot ❤️

[@TheAIGuy]

This is definitely a video I am working on. So stay tuned! Thanks for the suggestion and feedback. I appreciate it a lot!

[@osfree7612]

​@@TheAIGuy Can you make a video about how to run yolo as a web application using colab? Thanks for the videos.

[@chaitrakulkarni5034]

I must appreciate, this channel is so underrated. This is real gem. I have tried following your tutorial that's really working

[@karthikshet6226]

good job.please make a separate/ extended video for using custom obj detection on this same colab using webcam

[@jasurabdullayevabdullayev]

It was great I watched thousands of videos but still were facing errors but here everything is simple and understandable. Thankyou for the video and good luck !

[@chjayakrishnajk]

This is gonna be our new year gift

[@sanjanaprasad9093]

The one and only useful video in the entire internet !!! Thank you sooooo muchhh for the entire working code !!!!

[@AswinRamadasindian]

Right time at right place. You are my hero ❤️

[@TheAIGuy]

Appreciate it! Thanks for the kind words. Wishing you the best.

[@rohitjg]

I thought it is impossible to do it in cloud! Thanks For the video!:)

[@TheAIGuy]

So did I for a while! Took some research and testing but finally got it to work. Thanks for the comment :)

[@rohitjg]

@@TheAIGuy :) thank you for replying

[@mubinmodi7529]

Hey man appreciate your help and all the effort that you put in these videos, thank you 😍♥️

[@TheAIGuy]

Thanks for the kind words! Much appreciated!

[@taeefnajib]

You are an awesome trainer. Your topics are also just what I need at the moment. Do you have any tutorial on Object Detection from Webcam using custom datasets?

[@anshumaanphukan4388]

name 'Javascript' is not defined error shows up whenever I try to take a webcam screenshot

[@keitakeita7890]

wonderfull ! it is long time i look for to know how to use webcam on collab !

[@melontrophy7483]

Great video! Easy to understand and follow along the line video! For future video, would you be able to make a video about real-time people counting using this (webcam and collab if possible too)? That would be great and amazing!

[@TheCodingBug]

This was of great utility! Thank you.

[@TheAIGuy]

Glad it was helpful! Thanks for the feedback, much appreciated!

[@adeeshbhargava8136]

Thanks Guys , works brilliantly !! Easy to customize too , really helpful

[@arunaggiripandian154]

Bruhhh !! U are rocking ⚡ Please continue the same and don't forget the COLAB dude ❕ ... Keep it coming ⚡

[@hansgabrieldaduya2565]

Keep this up man. Good Job!

[@TheAIGuy]

Thanks! Much appreciated and will definitely keep them coming :)

[@xyzpatel8596]

@AIGuy please give solution for the problem in the object detection, it doesn't give error but also doesn't recognise the objects. Please help!!

[@karthikkp7867]

in 5:50 , when i runned the code it is not identyfying the person and object? what can be the error?

[@xyzpatel8596]

Yaa, mine too.

[@xyzpatel8596]

Please reply

[@yusufkenanturak2357]

Thanks, we are waiting for that

[@nachiketshirgur4066]

Thank you for providing us this tutorial

[@MegaMauricio120]

This is exactly what I needed

[@MaKrout]

Nice. I like your clip very much, it’s made me love coding.

[@yoshithaab7290]

I believe the code isn't able to perform object detection, it's just capturing picture! Please check it

[@bayuazra357]

yeah isnt working anymore

[@yoshithaab7290]

@@bayuazra357 can you please help in providing any other code for it, if possible!

[@TheJason143]

You’re a genius

[@devavratpro7061]

@The AI Guy, thanks for your great video and support. I tried running this offline but couldn't run Webcam part due to dependencies on Google colab libraries which are not accessible offline. In this regard, kindly guide how can we use js_eval function and others offline on CPU. Or kindly provide other Webcam integration method. Thanks again.

[@MrEri299]

excellent tutorial ... from Semarang Indonesia

[@Huds-ux1xb]

Bro ajarin

[@zilongli9070]

It's my huge pity for me to find you so so late. l am learning Yolov3 right now. Sending my best wishes from China. Be safe, l have shared your video to my Deep Learning Friends.

[@wwetop104u5]

Hi there this tutorial is very helpful for me Can you plz tell me what changes we have to done for external camera instead of webcam.

[@user-zy3yz2ju2f]

I am actually getting a few errors and while uploading the image and for real time the bounding box is not coming and the object is not detected. But my camera and all are getting integrated.

[@anandvihari3825]

Did you get the solution?

[@weareone2069]

Hey great video! But when I tried to run this ' network, class_names, class_colors = load_network("cfg/yolov4-csp.cfg", "cfg/coco.data", "yolov4-csp.weights")' I am getting this error- NameError: name 'load_network' is not defined. I am not able to solve the error

[@sharome90]

same problem here

[@andriasdaniswara7832]

@@sharome90did you find the solution?

[@easymatlab]

I wish you were my lecture you explained this concept flawlessly and I was following along with you ,eNkosi AI Guy (^◡^ )

[@nilakhidas257]

Amazing project indeed!!

[@iangraham6730]

Great tut! Thanks for sharing! Can I RTSP a security camera through this?? If so, how would I go about doing so? Thanks again.

[@GagarineYuri]

I am on this project. Don't hesitate if you want to colaborate. I achived to get the video stream from a security camera (Tap C310) on RTSP

[@Hitesh-Salgotra]

Wow this is amazing i was just wishing how i can train my model on webcam and there you are giving me gift for the new year..

[@TheAIGuy]

haha perfect timing. Hope you enjoy.

[@Hitesh-Salgotra]

@@TheAIGuy yeah i enjoyed a lot❤ can you make a video on scaled yolov4 using colab and tfod effecientdet on local machine as well as collab.. 😊

[@diwakarraju2976]

its a great content bro

[@TheAIGuy]

Thanks! Much appreciated. Glad you enjoy :)

[@OrionEx88]

Thank man, for this awesome video. It's really helpful for machine learning community. Anyway, I have tested with your code and it work perfectly, except the video streaming part. Mine was streaming with main camera instead of front-facing camera. Do you know which part I should change?

[@TechInventorAman]

how do we train it for our own cutsom data

[@shahidumar6513]

Was trying to figure out how can I access webcam on colab but do the detections with video stream! Thank u!❤️

[@TheAIGuy]

Glad i could help!

[@user-sp4fb2rj3o]

@@TheAIGuy Hi there, pls could you please let us know the answer to the question?

[@TheAIGuy]

@@user-sp4fb2rj3o what question are you referring to?

[@user-sp4fb2rj3o]

@@TheAIGuy if it is possible to do this implementation not with the camera but with a stream (e.g having a url). Well, to set this straight, I was wondering if it is so simple just like cv2 to put an address (http:.....) to capture function, in your JavaScript code?

[@ladieslinen382]

@@user-sp4fb2rj3o i have same question any one have answer of this question

[@FarukhKhalilov]

Hi. Thanks for sharing tutorial. Currently working on project and your tutorial helped me a lot. I have a question regarding selecting the cameras that are plugged in computer. As a default it runs my very first camera. How we can choose another camera? Thank you.

[@user-pr3nl7lv3s]

nice job. i have hard time makeing it.Thak you for teaching me makeing this.

[@OmkarJoshiV]

very nice example mate. do you have javascript client which you have shown in the example ? working on multi camera setup web app which can send stream to backend app(TCP socket) which holds YOLO and can detect objects if that makes sense. The setup would be remote camera streamed by Yolo on server side and transformed frames will be received by web app(client side).

[@xueyang112]

Thank you for your hard work. Can you give a tutorial about how to calculate mAP, precision and recall?

[@maninderramgharia1891]

no error in the code but still not detect any object #need support

[@devavratpro7061]

Hi, I have one request: I am working on an object detection problem where it is required to detect objects at two levels. At the first level, it has classes and at the second level, it has subclasses. For example, classes are birds or animals or flowers. And subclasses for animals are e.g., dog, cat or lion. For subclass level detection, it is required to add few extra CNN and Dense layers to the subclass level detection. In the present YOLOv4, I am not able to update the neural network layer architecture. Kindly suggest implementation. Thanks in advance.

[@ritikjaiswal9398]

Please give your model source code😂

[@Mlleshiin]

thanks for sharing these type of videos! they're so interesting, do you know how to change the input if we have more than one webcam?

[@anikdatta8055]

first view bro

[@TheAIGuy]

haha appreciate it! That was fast.

[@anikdatta8055]

@@TheAIGuy I'm following you from long time and you are always awesome

[@willcaster704]

These videos are amazing, as is your YOLO software. I know these are kinda old and don't know if you still monitor the comments, but I wanted to ask if it's possible to train the models to tag webcam images with specific people--like me, my family members, my dogs, etc. Thanks!

[@dbanuel3]

great videos!

[@andree912]

love this video

[@mamanghachi8086]

This is quality content, keep up the good work :) also is it possible to show how much fps generated in real time object detection?

[@alankarshukla4385]

yes

[@ucanhvu274]

Good job, thanks

[@SadatSafawi]

its giving error at the end of 100th iteration ): and gives this error : Saving weights to mydrive/yolov3/backup/yolov3_custom_last.weights Couldn't open file: mydrive/yolov3/backup/yolov3_custom_last.weights

[@cospace2113]

crazy video!

[@thetienkaka]

Thank for your sharing. I have a question. how many fps(frame per second) did you get when run on Colab using GPU? thanks.

[@shiffin_chippe]

Can I use YOLOv4 weights and cfg files from your old videos in this tutorial?

[@TheAIGuy]

Yes you absolutely can! It will work with any YOLO weights and cfg file.

[@shiffin_chippe]

@@TheAIGuy ❤️

[@andrezx10r]

I tried. It worked perfectly!

[@do0058]

Can you tell me how to upload other weights please

[@shiffin_chippe]

@@do0058 Just upload it to your google drive

[@posetexe]

great tutorial and content but i have questions. they tell yolo can perform 45+fps in videos but why is it so slow, like 15 fps?

[@gerhardheinzerling9880]

Thank you so much for the notebook. I started it and it ran directly without any errors. Very very cool. I think you are using coco. When I ran your yolo4 from your last video it directly detected a "sports ball". This time it never finds a sports ball but a "traffic light". Any ideas?

[@TheAIGuy]

You can swap back to the regular yolov4 cfg and weights file. The one used in this video is the scaled-yolov4 version which is faster and more accurate but maybe not for certain specific classes. So my suggestion would be to switch back to old weights if you found they were working better for you! Hope this helps.

[@gerhardheinzerling9880]

Yes that helped. :-) Thanks!!

[@andrezx10r]

Is there possible to do with custom dataset? I'm having trouble with the load_network() function for custom yolov4. I follow the steps and got the error: NameError: name 'load_network' is not defined

[@andrezx10r]

I figured out! It was the .data content. It is requested just the classes and names information.

[@TheAIGuy]

@@andrezx10r Awesome! Glad you figured it out and thanks for sharing the solution so others can hopefully do the same.

[@shwedhasrinivasan2833]

@@andrezx10r what to change the .data content into

[@andrezx10r]

@@shwedhasrinivasan2833 I commented the lines about train, valid and backup. Just used the lines about classes and names. But check the path of your names file.

[@shwedhasrinivasan2833]

@@andrezx10r thnks alot

[@LL-hr3bm]

Very nice :) How can you save the detected labels (only of a picture, not the video) in a dataframe? Thanks in advance!

[@ash3844]

HI, wonderful tutorials for beginners, but in this tutorial code can I increase few more testing objects, as it did not detect pen stand and batteries.

[@hadif6165]

Hi AI Guy! Do you have ideas on how do I need to add or make a change if I wanted to make a decision based on the detection. For example I want to display "It's Horse" on the bottom of the live camera when it detects the horse or any other object.

[@diegonovoa5901]

Hey Unable to init server: Could not connect: Connection refused getting this error when uploading a picture

[@rufamus7277]

Amazing content! Is it possible to use a live video feed from a Blackmagic UltraStudio Recorder instead of your webcam?

[@siddharthrana512]

camera 1 identifies two people, say Person A and Person B. Now, as these two people move from the coverage of camera 1 into the overlapping area of coverage of cam1 and cam2 and into the area where only cam2 covers, cam2 should be able to identify them as the same people A and B cam1 identified them as. i want help in that

[@congluong463]

thank you so much !

[@nielswitlox3539]

Hi AI guy, thanks for this video, it is very helpful. I am looking into ways to track the position of fish in an aquarium for a school project. I was wondering if there is a way to add objects to the AI?

[@braydonrainbow4909]

Hey mate, I'm trying to use this method but use the mixed reality capture of a Microsoft HoloLens instead of the webcam. I'm having issues using the URL of a streaming video through Colab. Any advice?

[@talhakamboh535]

Great video! Can we use this algorithm on a flying drone for object detection?

[@conor_codes]

Could you explain a bit more on how to upload your own model like at 4:21?

[@adarshsudarsan7503]

how to detect persons from real time video source( security cameras) and focus to persons for that need to adjust camera (pan and tilt) after that zoom to the person and capture the zoomed image and store image

[@kivlankhairwijayanto5404]

is there any way to give a minimum number of confidence on the bounding boxes? so its bbox appears when the 'confidence' value is above a certain number

[@dpredie]

is there a way to access video stream ( webcam, OBS) and apply a pre-trained model without using colab to output a close to realtime output

[@josephorimolade5626]

Good presentation, simplified enough to follow. Can anyone kindly explain to me the required modifications for running the code on a standalone window machine with Jupyter note; that is, not using google colab. Thank you

[@conor_codes]

Hey did you ever figure this out? I am running into a similar issue

[@carloselilopeztellez7802]

Hello! I'm using your yolov4 tf repo for video detection. Can you do batch detections? I'm running multiple cameras and want to process one frame per camera all at once. I've tried shapes (5,416,416,3) but got some errors. Thanks!!

[@LordWildbeast]

how to save video file like you save the captured photo? thanks

[@openmindjustdoit1306]

i have 400 employees they have many divaces like printer and computer and table manyhtinks they have so can i use this way to counters and save it in datset excel sheet or some thinks like that

[@user-nz3yd3eb3d]

hey man! I don't understand English, so I didn't understand what you said in the video. I tried to run your code. I understand correctly that he uses a web camera to recognize objects first in a photograph, and then in a video online? It's all?

[@aditinarendran]

I was trying out this code but during the webcam video part my webcam capture/output is filled with too many bounding boxes that i can’t see the image itself. Can you pls help.

[@_XY_]

Nice

[@zaheerkhalid5551]

How to change the camera port for this code? I want to use the different attached camera with my PC. I just don't know how to swap between cameras in the code.

[@Sakthi_22_04]

a great video!!! can you please done a video in real time emotion detection using google colab

[@mariomasseiro9090]

im running the code in google colab, but when i run the box for test the image with the guy and the horse i can't see any labels, also can take pictures and show my webcam, but dont show any label, u know what could be the problem?

[@Ippo_athuku_enna]

Awesome bro ❤️

[@Deshwal.mahesh]

How can you train your own Scaled Yolov4 if we're not using the pre trained weights?

[@MutenRoshiUI]

I have been trying to use my custom object detector using the webcam in colab but i am not able to detect anything, whereas i am able to run the custom detections on images and videos as shown in the previous videos. Can anyone help?

[@debarghyakundu908]

sir i have train model=253 mb size i wnt a code that acess my camera with model to detect sign language

[@Ksyusha5]

Good afternoon. tell me, is it possible to make a cloudy blender render in eevee? And how in this case to connect the active monitor?

[@amitpaul4848]

I created a project on object detection by yolov5 and without darknet.....but I can't add my local device webcam on it via google colab.....help me to connect my webcam in this colab

[@user-cc5ig3dg2r]

thank you very much

[@dsdummyaccount3213]

OSError Traceback (most recent call last) in () 1 # import darknet functions to perform object detections ----> 2 from darknet import * 3 # load in our YOLOv4 architecture network 4 network, class_names, class_colors = load_network("cfg/yolov4-csp.cfg", "cfg/coco.data", "yolov4-csp.weights") 5 width = network_width(network) how to resolve thsi problem???

[@ThienPham-ov6rv]

Same problem here. Can you fix it?

[@rob26r]

I’m a coin collector that would like to have a process that would identify a specific year or image in the coin. How would I go about that?

[@dwiyantosuabstian170]

please tell me how to make a custom dataset using a webcam using colab

[@prashantgupta1908]

How can i use it to train it for a custom object detector using webcam, i already have have done custom obejct detection using image by watching your video, i have weights and config file , but still not clear how to turn this into a custom object detector using webcam. If you could tell exacly what to change in this, it would be really helpful

[@krishnamote7647]

I am using this on my custom dataset. But I am getting an error: NameError: name 'load_net_custom' is not defined Would you please help me to resolve this error?

[@rohitscienceexpirement-tam639]

I tried with custom weights by changing the line in *import darknet** cell to to the custom files network, class_names, class_colors = load_network("cfg/yolov4-obj.cfg", "cfg/obj.data", "yolov4-obj_last.weights") and the notebook gets crashed for unknown reason? why is it happening??

[@andrezx10r]

Change your .data file. Use only the classes and names information! That worked for me!

[@rohitscienceexpirement-tam639]

@@andrezx10r i got it to work i just changed the coco.data and coco.names to my needs and it worked!

[@TheAIGuy]

@@rohitscienceexpirement-tam639 glad you got it to work. Congrats!

[@rohitscienceexpirement-tam639]

@@TheAIGuy 😄Thank you!

[@jihwanshin1233]

How can I customize the detection function so that I can count people on the webcam and detect people only?