Building Yoga AI Trainer using Deep Learning | Pose Detector | Tensorflow Movenet | TensorflowJS
Vložit
- čas přidán 9. 06. 2024
- Tensorflow Movenet Blog: blog.tensorflow.org/2021/05/n...
Tensorflow Pose Classification Blog: www.tensorflow.org/lite/tutor...
Live Project: eager-bardeen-e9f94f.netlify.app
Source Code: github.com/harshbhatt7585/Yog...
Twitter: / harshbhatt7585
Instagram: / harshbhatt7585
GutHub: github.com/harshbhatt7585
Thank You Guys for watching this video
#MadeWithTFJS - Věda a technologie
Good job Harsh.Spread the good knowledge to the world to make it better than ever.Good luck.
Super man...great project...loved it...absolutely great job... 🙏🙏
Thanks a lot man for this Great Video.
Brilant 🙏👍
great model, can you show from starting, how you trained and deployed the model from the vary beginning to end.
How hard is it to learn Keras for training the model and what would I need to do? do you know of any other neural networks that can be trained?
I am planning to make an exercise app, similar to this which detects how many repetitions you are doing
Hello, I can't find the executor from PC in the repository, so as not to use it in front. I read the comments and it says to use the classifier but it is not
Bhai git clone and npm install and then npm run dev karne pe local host refused to connect aa raha hai can you please tell how to run this project in much more detail, it would be great of you can make another video explaining the implementation part in more detail as I am not understanding how the model is determining the correctness of the posture as movenet only gives the key points
Thanks for embracing TensorFlowJS and Movenet!
I was able to train the model . how to predict on 1 image and print the result? how can i convert from images to coordinates? please help me !
Bro i have final year project based on this can you help me build it in react native!!
Good explanation ! I trained my dataset over coded provided by u and saved model in group1-shard1of1.bin and model.json file but when i am connecting the model.js instead of yours it not giving answer. More onwards when i used your local model.js file which provided by u instead of s3 bucket's link provided by you. Would plz tell me what im i missing
how much time did you take to train the model, how many epochs??????
Hi harsh ,
Thanks for the tutorial.
I am applying movenet on videos to extract features from them .
Does it ignore some of frames in a given video ?
Basically movenet extract features and predict the keypoints around 20 to 24 fps in browser even more if you have better gpu. And in terms of ignoring frames it depends on you how much frames you are giving the model, at the end you are just giving it photo(1 frame) and it predicts it in some milliseconds and in a second it can predict 20 to 24 frames
in training file I'm encountering this error, can you help me with this AttributeError: module 'numpy' has no attribute 'object'.
I want to go to the next pose automatically if I did the pose in 5 seconds. please tell me how to do it
I was very impressed. I would appreciate it if you could answer one question.
1. In github source code, can the "classification model" code be executed in Windows operating system?
Absolutely
can u teach us how to build these project , yoga
Bro let me know how to run this I can't able to run give a description in GitHub that will be helpfull for run. I don't know how to run these python and js with combined. I'm getting errors while running js home file. Please let me know the process at all about the project.
very very good explaination
Hey Great video, I have a question, what would be the best way to deploy this and similar pose detection models to the internet ? what would be the best framework and also best way to open the webcam? thanks
Sorry for late response, the best way to deploy this using TensorflowJS for now as it supports browser and uses WebGL also, the framework and library is the same if you are using react you can able to deploy this using TensorflowJS and even you can use react native to tun this model in mobile phones. There are lots of ways to use webcam like I used React so I installed a webcam library and similarly if you are using other frameworks or language like in python you can use OpenCV to use webcam, there lots of good ways but remember it should be giving you every frames then you can able to inference the model
Bhaiya please python tutorial dona basic se 😘. Comment from Tanay bhatt in mano phone
Could not find the dataset please give the link for dataset?
Could you please tell us "What is the pros and cons of this topic"
Hello bro could you please tell me how you have deployed this project and which tools or technologies or libraries or websites have you used ?
plz tell what are the libraies
Anyone added new poses?
Jarvis tutorial please 🙏
Could you please tell briefly about deployment of model.json file
i followed your tutorial and implemented every thing its working for test and train csv files but i have another application where i have a video it consists of different poses i am dividing the video and send it to it but is seems it is predicting single image pose values seems , i am also using the pose embedding which is in the human pose tutorial can you suggest me how to predict that single pose with in the video, Thank you
can you explain how did he caluclated final dataset from keypoints dataset
plz share link git link for your work too
Hi thanks a lot for the explanation
can i use it to detect more than one person ?
Yaa you can do that also but you have take different model which is multi pose detection, this one which I used is single pose detection, you can find the model in the tensorflow and mediapipe
excellent work !! can u plz tell me from where you got the dataset?
some of them I got free dataset from the internet and some I downloaded the images of each class, If you want to use the dataset, I have uploaded in my GitHub you will find the link in the description
Hi can u show step by step on how to it plis :(
How do you convert as a live project link?
how can i add more pictures of asanas to your code?
hey bro loved your video
i am also doing pose correction system and i wanted to ask you how did you convert coodinates of keypoints into csv file
It's easy, I use cvs and pandas module for converting into cvs and reading csv file
I m Also working on yoga pose correction in real time.can we connect
bro i immediatly need project can u give me urs
Hey I need your help! i want to implement this! but couldnt find directions! can we connect so that you could help me!
Hey if you got can you help me with directions here
Hi iam divyanshu ❤️
is there any specific higher hardware requirements when testing it in real-time? pls reply
yes it depends on the system requirement and GPU to work properly
Can you provide the frontend connectivity code to backend code... Because the code you provided on GitHub is directly executable by frontend using tensorflowjs .... So can you provide frontend code that can be connected to backend..
I think backend you mean here training model and deployment. There is no direct connectivity of it but you can generate model and deploy it seperetly then use it in the frontend.
Wich is btr for key point data movenet or open pose and why. What are the differences
is your code public? and for real time testing did u used specific hardware? pls answer
can u share ur email or whatsapp for communication
Hello harsh brother , I am facing issue , "missing destination file operand after npm or node.js"
can you say more about the issue?
Can we run the code directly without help of Front end (node js and npm)
Yes you run pose estimation directly in python
Which nural network u used in your model
I used Linear layers to train a classification model
what have you used to create the frontend?
ReactJS
hi bro, your web app isnt detecting yoga pose after opening live project that you've deployed on vercel.
from my end it's working. please try again some time
@@harshbhatt7585 It worked for me too later thanks!
Can you help me, I can't run the project
@@bhushanambhore8378how did you run the project could you please help me out
Can you please explain how to execute this project after getting code from your github?
You can use my classification model or you can use my code to train your poses what you what you want to classifying, if you want to make classification model of your poses you can put your training imges in the "classification model/yoga_poses" and run the file "classification model/training.py", it will generate model.json which you can use it in javascript, well you have to host this model somewhere in cloud and then you can fetch the model through tensorflowjs and use it, you can use it same code of mine with some minor changes like replace the old pose names with your pose names.
I think this is more looking complicated but if you want to know more feel free to get back, I will try to help you
@@harshbhatt7585 Thank you! It’s done but the tree pose is not accurate. It shows green skeleton even when the leg is below the knee. Could you please tell how can I correct this?
@@srushti1258 you can correct this by adding another feature which is angle between joints so if angle between leg and knee is not 180 degree then it can predict accurate. it will also help in another joints as well like hands and head position
I need Jarvis tutorial please
🤣🤣
Can you please tell how to learn this project from scrach as I have very little knowledge about ai . Any course please to make this project in less duration of time???
I have given some blogs in discription for further reading you can take a look at it.
All you need to know to use CZcams, Wikipedia and Twitter to learn any stuff.
bro how do you run it
You can clone the project and then run it using ReactJS
Which version of npm??
it will work above npm version 8
hey can you tell how to train model to add more yoga poses
You can use my classification folder on GitHub and add more images on my datasets then run the train.py you can generate the model
@@harshbhatt7585 I meant to say to , add different yoga poses apart from the ones that were already present. After training the model csv file for newly added yoga pose is not created.
@@amanpreetsingh-ty5xo you can put the new images into same yoga pose folder and if still csv file is not generated, you can check on the preprocessing file the name of the folder and change it your folder path/name
Wow! Amazing video!
Im working on my final project (which us directly relationated with your project) t o finish my degree of Electronic engineering.
Can we talk a bit further? I really need that
See u!
Thank you, Sure we can talk
Bro, i want help from u on this project so can I talk with you?if yes let me know ur contact no./social media account.
I wantto make it in Android app is it possible?
Yes it's possible, you can use React Native to create IOS and Android Application by using JavaScript
Hey bro!
I tried your project Yoga AI trainer and it had worked perfectly in my pc.
I wanted to try extending the amount of poses so that i could learn how it worked and it didn't properly work when I tried so.
I added the test and train images for new poses
Then I tried running preprocessing file and then training file
The csv files for the newly added poses weren't being created.
Could you help me out?
Could you please help to where to learn this project ? Any course??
how to run bro ??
@@kunalpatil1910 npm install
Npm start
@@bingoovic3676 Hello bro, i dong it that, but it isn't work
Send the code to me to bro,I don't know how to run
Git repo ?
github.com/harshbhatt7585/YogaIntelliJ
Sabhi tarah key parabola ki universe🌌 level par solution👌 nikalva do phir sara universe🌌🌌🌌 ka solution👌 ho jayega
can u provide ur contact details, I am looking for implementation of yoga pose detection and feedback model
in the about section, you will find my social link, feel free to connect over there.
It's good project please send me all the code
it’s already uploaded in the github, please checkout the description
kindly send your contact no I want an ai enabled self check for sun salutation steps for kids