AI Pose Estimation with Python and MediaPipe | Plus AI Gym Tracker Project

[@winstonyeung1461]

Sir, your presentation is really systematic and content is as always informative. Thank you so much.

[@NicholasRenotte]

Thank you so much @Winston, glad you enjoyed it!

[@shukkkursabzaliev1730]

Mind blowing tutorial! Look forward for more similar projects, especially the one you talked about (counter for several exercises at once). Liked + subscribed + bell turned on!

[@gnm81]

When will that video come out

[@sikandarencosa2490]

this channel is such a gem tbh

[@NicholasRenotte]

Thanks so much @Sikander encosa, pumped you're enjoying it!

[@MJ720]

I decided to try and use mediapipe for a project I was doing but got discouraged by some of the documentation as I struggled to even install it. I then came to this tutorial and you made it look so easy and understandable! Thanks.

[@forgewhelbon1131]

Did u get any os error trying to install media pipe?if so how did u fix it?

[@MJ720]

@@forgewhelbon1131 I didn't get any errors sorry. Good luck in sorting!

[@forgewhelbon1131]

@MJ720 oh ok :( thnx tho

[@LabbHasse]

Thank you so much. Your teaching and tutorial style is the perfect match for me. Clear an precise, and you continuously explain everything you do, thanx!!

[@minhnamphan864]

I needed to calculate the angle of my arm to my body and your video helped me a lot. Thank you very much. I hope you release more videos

[@mathieuraffinot1126]

Many thanks for this tutorial ! The video is well done, your explanations are very clear, and the code ran directly on my computer without any modifications, which is quite rare !

[@bankcrawpackchannel6936]

Please make more MediaPipe hands tracking projects, we love this content thank you so much

[@NicholasRenotte]

Oh you know it! Plenty more coming @Bank Crawpack Channel!

[@bankcrawpackchannel6936]

@@NicholasRenotte thank you so much

[@epidemiaproductions]

I've saved this vid for later, dude you are just amazing, i'm loving all the vids and the way you teach is really easy to understand, keep it up! cheers from brazil :)

[@NicholasRenotte]

Thanks so much @zanniboni! Much love!

[@bhaskarkc7276]

I am learning python and its been week but i can totally follow you. Great work.

[@jerrylee520]

This is the best video I saw to use & test pose estimate. Thank you.

[@muhammadmemon2394]

Excellent step-by-step video! Great work! So much help for my capstone!

[@AshishPandey-cp7dm]

and once again super thanks for doing this. You make it super easy for a beginner to understand stuffs. This was your second tutorial i tried (1st one being Object detection - 5hr course).

[@ishanubale9680]

very good video all info from the start and no complicated installations

[@NicholasRenotte]

Thanks so much @Ishan, tried to keep it as smooth as possible!

[@madebymate4870]

Thank you dude! You are the most underrated youtuber I have ever seen I'm loving this video although I have one question How do you do the exact same thing But instead of a camera, You do it with screen recording live? Hope you or anyone else can help me Cuz I'm really stuck

[@NicholasRenotte]

Could access a screen using pyautogui and use that as the video instead of the feed from opencv. Check this out www.thepythoncode.com/article/make-screen-recorder-python

[@madebymate4870]

@@NicholasRenotte thanks!

[@udaysodhi7574]

More content creators like you should exist

[@파란색동물]

I copied your code, but the angle is not displayed on the screen. I don't know what the problem is.

[@ipirediatoir247]

if you found a solution please tell me

[@kaviyak7308]

Did you find a solution? I'm dealing with the same problem

[@rafaychaudry320]

@@kaviyak7308 did you find a solution?

[@leamon9024]

Thanks. This is amazing. Looking forward to the multi-pose tracker tutorial.

[@captainlennyjapan27]

I just tweaked this to track my basketball shooting motion in just an hour!!! Thank you for the awesome tutorial 👍🏻👍🏻👍🏻

[@favourchianumba5823]

hey bro could you send me your github, id love to see how you did it, Im having trouble implementing the first block of code after I import everything and Im not sure what I am doing wrong

[@stokrandy]

I’m interested in how/what you did. The reason I’m watching this is track my daughter’s softball fast pitch. The initial idea was to identify what pattern she uses to get the strike. I thought I could monitor her movement, read the speed of the ball, and determine if it was in the strike zone. Man, was I kidding myself. lol there is a lot to each of those tasks. Too much to accomplish all at once.

[@Flipszn]

THANK YOU SO MUCH. YOU SAVED MY SENIOR DESIGN PROJECT

[@NicholasRenotte]

Ayyyyy, awesome work my guy!

[@ademhilmibozkurt7085]

What a video Nich. I love data science and computer vision contents. Thank you

[@abhishek_maity]

After MediaPipe Please do cover YOLO v5 /v4 ! I love your courses. They are really awesome! :)

[@manishsharma2211]

Namaste sir

[@abhishek_maity]

😂😂😂❤️❤️

[@NicholasRenotte]

Ayyyeee, you got it!

[@ancycoding]

Really powerful library of python and the teacher too.

[@sangitaajay3266]

Cool project Nick you are such a great coder

[@NicholasRenotte]

Thank you so much @Sangita, much appreciated!

[@mikecooper8142]

here for the youtube algorithms. I will definitely watch it later tho. :)

[@NicholasRenotte]

Thanks so much @Mike, let me know what you think!

[@khronos142]

mythmon spotted :3

[@sankets7501]

I am studying data science here in mumbai, India. I've got to make a capstone project and this video inspired me to build a home workout app, still working on it. Thank you so much mate for all these videos, you are a HERO. One quick question : should i use movenet lightning or mediapipe pose shown in this video?

[@cattnation6257]

what do you want

[@cattnation6257]

go with mediapipe

[@malekaburaddaha5910]

Man you're the greatest, this is just what I was looking for.

[@khadijasharifi4814]

Hi, I really appreciate the effort you put in to break everything down and make it understandable. I want to ask a questions; what does 'image.flags.writeable = False' do exactly? I understand that it improves performance but how does it do this? Many thanks :)

[@upenyuhlangabeza7467]

Interested in knowing also

[@crownsandcapes6101]

Any advice on how someone would use this method to track treadmill steps? Awesome video, BTW, very engaging!

[@henkhbit5748]

Super cool and awesome Nick!.

[@NicholasRenotte]

Thanks so much @Henk!

[@DJPapzin]

Great tutorial, you are such a great teacher.

[@KJletMEhandelit]

Great video, I'm happy that i learnt something via implementing with proper guidance.

[@williambittner8828]

Completed the tutorial and learned a looot!! Thank you very much :)

[@aleynahukmet5996]

Hi Nicholas! First of all thanks for amazing tutorial. It was really helpful. I love your tutorials so much you are explaining the import details in very understanding way. But I want to ask you a question. I did not really fully understand the angle calculation part and I could not find anything about it on the internet either. Could you please give a little bit more info about that part?

[@NicholasRenotte]

Definitely, take a look at this: manivannan-ai.medium.com/find-the-angle-between-three-points-from-2d-using-python-348c513e2cd

[@fahnub]

Super Project. Thanks for creating this @Nicolas

[@manudewi]

Thankyou so much for the tutorial including the notebook man!

[@ramiro6322]

First of all this channel is amazing, I just found it! Second of all, I have a question about the angle calculation. How's that it's independent from the camera angle? You're using projected 2d points to calculate an angle that actually exists in a 3d space, so the camera angle can change the end result a lot. Am I missing something? Anyway thanks for the great content

[@NicholasRenotte]

Nope, you're correct, I simplified it by calculating it as though it were in a 2d space. You could refine the calc by adding depth to the angle formula as well though!

[@Ramix09]

@@NicholasRenotte thanks for answering, I'll have that in mind for my project!

[@harshpatil6168]

@@NicholasRenotte Can you provide an example of how to consider depth as well in the angle calculation?

[@Ramix09]

@@harshpatil6168 I think you'll have to use another camera or use a library that also has 3D predictions even with one camera

[@陈逸飞-l8k]

@@harshpatil6168 Using kinect camera can solve this, it can detect depth

[@jackluk5471]

As a beginner coder, I found this tutorial helpful and very easy to understand. There is only one part which I can not seem to grasp. I have gotten the code correct and even tried copy and pasting the one in the GitHub, but I am unable to get the angle to display on my screen. My assumption is that I have gotten the webcam dimensions incorrect (which you put as 640, 480), and I am not sure how to find out what the correct number should be.

[@parinayseth8813]

Did you find the solution for it ??

[@nilutpolkashyap]

Yeah......this is great.....was waiting for this project

[@NicholasRenotte]

Awesome!! Stay tuned, another sweet one coming on Sunday!

[@chetanmadhok1221]

Well explained. From Start to End.

[@utkarshsrivastava9241]

Thank you so much Nicholas for this video, learned a lot from it.

[@zindamayat]

Bro, you are an angel for me. I have the same FYP Body Posture detection correct count in my Gym App. You save me. Thanks, Bro.

[@florianbouron8340]

Neat, I’m doing the same as well haha! Love his videos he is such a great teacher! @Muhammad Mehmaam did you open sources your code by any chance?

[@zindamayat]

It depends on my colleagues, but I will try to upload in github.

[@florianbouron8340]

@@zindamayat Thank you so much

[@berkay3148]

Were you able to implement other workout movements?

[@mustafasaidkorkmaz8838]

MY MAN YOU ARE THE REAL ONE!!!

[@kusumabhargavinadendla]

Very very helpful. Glad I found it. Thank you ❤.

[@sorochinsky]

So much new information... Nick, I want to sleep at night😂😂😂😂

[@NicholasRenotte]

😂 wait til you see what's coming this weekend 😉

[@sorochinsky]

@@NicholasRenotte Oh my God😱 Scary to imagine. I'm looking forward to it.😁😁

[@NicholasRenotte]

@@sorochinsky yesss! Think I just wrapped up the code for it this morning!

[@sorochinsky]

@@NicholasRenotte This is very good, I just finished with it😁😁 I am infinitely happy😁😁 Now I can get the coordinates of the points. It is very cool. How many things you can think of.

[@NicholasRenotte]

@@sorochinsky yesss! Awesome work. I've got soooo many ideas planned for it!

[@Gabbosauro]

I will keep an eye on this since it will somehow become my master thesis! :)

[@NicholasRenotte]

Awesome stuff @Gabbosaur! What's your thesis on?

[@Gabbosauro]

@@NicholasRenotte Basically it's a sort of personal trainer but more focused on calisthenics workout types. It checks the quality of your movement and re-schedules the workout program based on your performance.

[@NicholasRenotte]

@@Gabbosauro nice! Sounds awesome man!

[@aaronwee5956]

This is amazing!!!! Do more videos of this pleaseee

[@tsangwingho2508]

I love it very much! Is it possible to calculate how accurate our poses are in order to make sure that our poses are correct comparing with gym trainer ? Thanks!!

[@NicholasRenotte]

Definitely, I'm working on a video on that rn @Tsang Wing Ho! Stay tuned!

[@mxolisinkosinathinene2657]

Love your work it’s always on point and it teaches a lot for me as a beginner

[@NicholasRenotte]

Thanks so much @Mxolisi, super glad you enjoyed it!

[@aliriano15]

Awesome!!! Very interesting with lots of applications!

[@NicholasRenotte]

Agreed! Tons of stuff you can do with it!

[@amanpreetchander7386]

Really amazing tutorial. Keep it up. It really helped me a lot. Thank you 🙂

[@lushankshambharkar124]

Hello? Did you get the result?

[@venkatr9052]

May I know which webcam you are using. Btw, love your content

[@NicholasRenotte]

Thanks a mil @Venkat, I'm using a Logitech StreamCam

[@marwamahmood8434]

omg love your Video thanks and like the way you code it is so clear

[@МаксимОрлюк-л3щ]

Amazing job, man! Keep it up!)

[@ggggdyeye]

excellent video. i learned a lot of things about ML in this video Thank You so much.

[@softsilksarees7991]

Amazing video Nicholas I was really amazed by your video ai gym tracker. What are all the workouts does this system could recognize or it measures using the position of coordinates.

[@anuncibayz]

Excellent! Thanks Nicholas!

[@osmiumhelium]

Your all tutorials are so amazing and productive. In this lesson, I want to just display the angle in a whole number, not a integer. How can we do this. By the way thankyou for creating a nice tutorial

[@alifranakmalrahmadyo7592]

it helps me a lot. cheers from Indonesia!

[@taimoorneutron2940]

great job and made very easily code so any one can learn quickly

[@tawiwutcharuwat8999]

Great content, loving them

[@abhijeetsaraf7887]

Dude you are awesome. !!! Did the multi joint estimation video out, if its there cant find it out !!! I wish to make a yoga pose detection... Thanks in advance

[@fatimah-hussain]

awesome! quick question: how can I compare two different forms, one from a workout video and one from a person trying out their exercise?

[@thewhiteheart4439]

THIS TUTORIAL REALLY WORKS I AM FROM PHILIPP

[@Uncle_Buchi]

YES YES YES!!! The video I more than needed. So helpful! Thank you!!! If i wanted to use this method to calculate work done (WD= Force x Distance), how would you suggest I go about it. My thinking is that I'd assign a certain weight to either the joints or the line between joints and use that for force (signifying weight distribution in the body) then the model tracks the distance each joint or line moves throught space and multiplies to get work done. The individual work done per joint or line is summed and I'd have total work done by the body. Practicalizing this is what I can't seem to figure out.

[@NicholasRenotte]

I actually looked into this a while ago, it'd probably be something along the lines of angular force. Take an assumption of the mass of the arm, then calculate it's accleration (speed between frames) then assuming you know th length of the arm or use a proxy e.g. coordinate to coordinate euclidian distance...you should be able to do it!

[@arnavmehta3669]

You want to use that work done found as an estimate to Joules of energy used?

[@gnm81]

Hey you mentioned you will you a multi exercise tracker in another video, have you done that already?

[@azinkey3963]

Pretty good, my search end here

[@kanyinsolaolajide648]

Hello Nicholas, I was wondering, is the accuracy of the angle affected by depth? or by the angle from which the camera is capturing from?

[@manudewi]

It is affected by depth, just run the code and keep your elbow at 90 degrees as you turn around your shoulder. With 2D input you get a 2D angle! Edit: An easier way to trick the counter is to picture the curl with your forearm towards the camera so the segment lines between shoulder-elbow and elbow-wrist are in one line. Then just move the wrist over and under the elbow and you get a change between 0 and 180 degrees. Conclusion: For accuracy you need to film from the side.

[@Mrsimanater]

this was a fun video cheers nick:)

[@LinhPham-vg3ud]

great, can you combine mediapipe with yolo to detect multiple objects in 1 camera, i'm looking forward to it!

[@SaberSotoodeh-g3z]

Thanks for your great tutorials. Is there any tutorial, that shows how we can customize/modify or retrain the Mediapipe models? Current models are trained for adult pose detection and have issues in pose detection in infants. How can I improve it?

[@kanyinsolaolajide648]

Hello Nicholas, Great work doing all of these; it was enlightening. I wanted to know if you have done the multi-pose/ multi gym tracking you spoke of around 52:11. Thank you.

[@NicholasRenotte]

There's a vid on the channel somewhere, I just did the estimation not the pose detection!

[@kanyinsolaolajide648]

@@NicholasRenotte Thank you so much! Is there a way to estimate if specific key points in a video fed into the MediaPipe are missing? Say a video of me with just my head and shoulders, and I want to calculate the key point of my ankle. Is this possible?

[@manisharya2138]

Big salute for you work 🤩🤩

[@karamvirbindra9050]

@Nicolas Thanks it is such great help. I have 2 quick questions >>>> why have you multiplied in angle code np.multiply(elbow, [640, 480]).astype(int) >>>> how to resize the video frame if you can suggest some learning sources on same. It will be wonderful. Big thanks again man

[@NicholasRenotte]

Heya @Karamvir, 1. This is because we receive normalized landmarks we need to return them to our frame size to render them based on our baseline image 2. Can resize it by adding cv2.resize before the render :)!

[@rahulkumarpal385]

@@NicholasRenotte can you give the landmark for all body part such as knee , shoulder to render in our frame

[@felixmariochrisantoadi5850]

thanks a lot nick, i succeed follow the whole tutorial, and also add some more function lol just kidding haha i only manage to adding the other hands to working, so now i get two pairs of strong hands lol thanks a lot nick for ur help :)

[@CodewithLakshit]

Man thank you so much for the wonderful work!

[@suntrakaneshsu1997]

Really Awesome !

[@kowsalyav4159]

Hye Nicholas!! Nice work man!! I have a doubt. What are the changes to be made in the code when two cameras are used?

[@aryendrasingh4401]

First a fall great experience of learning from you. My question is why didn't you use 'z' axis for calculating angle? as in case of predicting body pose it may affect the accuracy of angle.

[@NicholasRenotte]

Don't really need it in this case @Aryendra as we're just calculating in a two dimensional space. You could try applying however!

[@AyoubBenayache]

Bad results with z. You should choose a good position to get correct calculation

[@estherlisiane9786]

Amazing, does this work with depth cameras?

[@farzanevafaii5820]

thanks for this tutorial! just one question ... what if we want to detect more than one person?

[@nakshatrasingh7853]

multi-gym tracking video soon? great content

[@NicholasRenotte]

Multi agent?

[@nakshatrasingh7853]

@@NicholasRenotte yes, when will that video be uploaded?

[@NicholasRenotte]

@@nakshatrasingh7853 hmmm, probably a while away, haven't tested it out yet!

[@nakshatrasingh7853]

@@NicholasRenotte great man, keep up the work

[@Alex-gu4mi]

Hi! Great video. I have a question, can you do the same about calculating angles but with the video of holistic with hands and face detection? What things should I change to make it work? Thanks!

[@NicholasRenotte]

Just need to change the keypoints you pick up! Planning a bigger video on rep counting using ML as well!

[@zabeelzabeel5856]

Hey, great video. Is this the BlazePose model?

[@hodglotser6505]

Great Video Nicholas! Thank you very much. I have just one question- I am having a problem installing mediapipe. I used "pip install mediapipe" but it didn't recognize the library. Would appreciate the help. It's kinda urgent.😄

[@NicholasRenotte]

Shoot, might be your Python version, can you try 3.7.3 instead

[@codaemon375]

Great video! Why use angles for the curl tracker tho? It would be easier to use the position of the wrist and elbow. If the wrist’s position is higher than the elbow’s position that counts as a curl.

[@NicholasRenotte]

Could definitely do that @Paul, wanted to showcase an alternate technique. Could also just ML to predict top of the curl and the bottom like in my gym video.

[@codaemon375]

@@NicholasRenotte thanks for clarifying

[@Kishi1969]

Great lecturer

[@NicholasRenotte]

Just trying to spread the coding love @Benya, glad you enjoyed it!

[@CiaoGhost98]

Great tutorial...how do I add more exercises ?? Is there any video about it?

[@NicholasRenotte]

Not yet, working on it my guy!

[@CiaoGhost98]

@@NicholasRenotte thanks

[@wenhan123]

Woah mind-blowing tutorial, may I know, if I want to use this project for my home workout routine, I want to add in more, such as jumping jacks you know any idea how? Like the codes start by first doing 10 jumping jacks rest 10 sec then the next workout will begin ( which in this case, is the dumbbell curl)

[@IaconoAndre]

Loved the video! Thanks for sharing bro :)

[@NicholasRenotte]

Thanks a ton @Andre 🙏

[@SaiYennu]

thank you so much information learned and talking about the tutorial display with angle only ,why not displaying time beside the top corner of the output

[@ElectroCoderEC]

Nice tutorial!. Great!. Do you know if you can use this for multiple dector pose for people at the same time? How can i get it?

[@NicholasRenotte]

Heya @TechTales SC, this model doesn't support multiple people. Check out OpenPose for it

[@newtonsarr1234]

Very cool project. Have you done a pushup counter yet ?

[@NicholasRenotte]

Not yet!

[@abdurRahman-od3fd]

Is it possible to add code for velocity between each movement of the dumbbell curl? thanks!

[@ozguozkan9342]

Hi thanks a lot for this amazing video! I want to ask something, do we use a tensorflow model or a tensorflow lite model?

[@vaishnavimore754]

Thank you Sir!!! For such an awesome content🤠

[@NicholasRenotte]

Thank you so much @Vaishnavi!

[@richieithaca]

Hi Nick - another awesome working example - is there a way of calling on the angle calculation script you use in this example without having to write out the code for every angle over and over (if I wanted to display hip, knee, and ankle angles all at the same time) - Also is there a way of rounding up the numbers to 2 decimals? Feel like I should be paying you for this level of assistance but appreciate what you're doing here

[@NicholasRenotte]

Yup! Stay tuned, might try to cover it this weekend or early next @Richard!

[@drm8164]

@@NicholasRenotte PLEASE HELP : I paste to script in PyCharm but nothing works, i have no errors but i can only see my face in the webcam

[@Neojivago]

"Make Detections" and the parts after that doesn't work. I run the codes but nothing happens. Do you have any idea about that? Could you please help me? Thank you so much.

[@yongbinlee9698]

Thank you!!, Your video is really helpfull to me, When I make pushup counter, I subscribe you and push the good button!

[@priyambordoloi771]

Nicholas by default open cv use cpu? So are you using any gpu power to increase fps?

[@NicholasRenotte]

The OpenCV bit should be pretty quick. Double check you're using a GPU for the MediaPipe component however: google.github.io/mediapipe/getting_started/gpu_support.html

[@priyambordoloi771]

@@NicholasRenotte you are amazing.