I Tried Teaching AI Sign Language

[@krishshah5844]

Always comes up with the content i need.

[@NicholasRenotte]

🙏

[@spencergarth5787]

lol "Frequently talks to himself". Grade A content mate!

[@NicholasRenotte]

🤣 frequently is probably an understatement. Thanks @Spenny!

[@nigaralishzade2379]

Thanks so much! Consider I want to make a dictionary with grabbed labels of shown signs in order they have shown. The sequence of labels: letters, words, and combinations of words accordingly to make up words, sentences, and texts from signs. What will be like this code block in Python? What is the variable I have to work with?

[@NicholasRenotte]

Heya Nigar, there's quite a bit involved to handle the words. Want me to make a video on it? I could do it as a scissors paper rock game or something similar so you can see the detection history?

[@nigaralishzade2379]

@@NicholasRenotte You are amazing sir🤗 I want to collect detection history

[@NicholasRenotte]

@@nigaralishzade2379 ohh, thanks so much!!! Added to the list! Might just do a simple example to begin with!

[@abisa5268]

Great content brotha, but how to detect if the gesture is a hand waving up and down or rubbing my hand together?

[@NicholasRenotte]

Heya @Abisa, you can use another technique called action detection! Vid on it coming soon!

[@dcryan]

There should be a box & label around the "Peace" symbol at the very end 😂

[@NicholasRenotte]

😂 ohhh man, such a missed op 😂😂

[@fauziramdhani6804]

hy nicholas thanks for your amazing tutorials, may i ask why im stuck on opencv imwrite not working and videocapture always freeze in the end? hope get the answers thanks

[@WeiQin-n9y]

Hello, could I know is it possible to translate sign-language with gesture motion???

[@NicholasRenotte]

Yep, there's a technique called action detection that should allow you to do it! Working on it as we speak!

[@WeiQin-n9y]

@@NicholasRenotte How to use the technique you said? Have any video to refer?

[@NicholasRenotte]

@@WeiQin-n9y I don't have anything yet but I've got something planned!

[@isaackay5887]

Exactly what I was thinking

[@NicholasRenotte]

@@isaackay5887 yup, can be done using action detection. On the roadmap to get it out there!

[@ayhanxd8609]

What is the difference between this video and full tutorial?

[@NicholasRenotte]

A little less technical in this one, this shows the capabilities without going deep down into the coding behind it!

[@ayhanxd8609]

Am i see wrong or you made 200k steps? It takes 26.6 hours on my pc 😂 And i still have same problem. When i add more than 5 words my accuracy totally broken. If i train with more steps can i solve this problem?

[@NicholasRenotte]

@@ayhanxd8609 woah, are you training with a GPU? How many images and labelled objects do you have?

[@ayhanxd8609]

@@NicholasRenotte I have gtx1050Ti. I dont know what am i using. How can i learn? And i have 150 pics.

[@NicholasRenotte]

@@ayhanxd8609 run through this: kzbin.info/aero/PLgNJO2hghbmhUeJuv7PyVYgzhlgt2TcSr

[@kanwargauravpaul]

Loved it

[@tushitasharma4860]

What is the detected variable for word?? I was thinking of converting it to sound. But none of the variable is coming out to be the detected word.

[@NicholasRenotte]

Heya @Tushita, check your other comment, it's inside of the detections object but you need to map it to the key.

[@ikf937]

how to use to hands by visual studio code ?

[@flaviolivieri]

Hy Nicholas, once finished, does not show me anything (boxes, names etc ...), the webcam work, but don't recognize. Please help me

[@NicholasRenotte]

Definitely, check my other comment, let's start debugging!

[@flaviolivieri]

@@NicholasRenotte thanks, I replied

[@NicholasRenotte]

@@flaviolivieri perfect, will take a look!

[@kirtipandya4618]

Which software did you use to create box and label?

[@NicholasRenotte]

LabelImg, super easy to use: github.com/tzutalin/labelImg

[@jiteshaneja3892]

Hi, Can we run Object detection more than one time on the same data-set, like: for the first time my object detection algo. will detect the only person in detection and in second step it will detect either the person in the last detection is wearing a mask or cap? if Yes then can u please help with how??

[@NicholasRenotte]

Yep, you would need two models, the first would detect the region of interest i.e. the person then the second would use that region to detect if the person is wearing a mask!

[@jiteshaneja3892]

@@NicholasRenotte Thanks for the reply! Can u tell me some resources where I can learn to do that as I am a beginner in this AI field and want to work on this project? It would be great help

[@NicholasRenotte]

@@jiteshaneja3892 that particular technique is quite advanced. Possibly start out with this: kzbin.info/www/bejne/f4CsYaJpeN2pb7M

[@jiteshaneja3892]

@@NicholasRenottewith help of this tutorial only , I completed with this ,I have two trained models with specific detection I just want to know how can I Combined the detection as output of 1st given as input for another

[@NicholasRenotte]

@@jiteshaneja3892 will add it to my list of tutorials, it's probably a bit too long to describe in a comment.

[@ramzeye]

What are the AI techniques used in this project ? Plzz tell me

[@NicholasRenotte]

Heya @Ramiya, it's using the field of AI called Deep Learning and we're applying a technique called Object Detection to do it. The Deep Learning model used is called a Single Shot Detector.

[@truptinandanwar5543]

Can we do these in cpu

[@NicholasRenotte]

Sure can, it's just a little slower when training!

[@weebotzu5533]

While training the model I'm training this on CPU ImportError : cannot import name 'model_lib_v2' from 'object_detection' (C:\Users\Lenovo\AppData\Local\Programs\Python\Python38\lib\site-packages\object_detection\__init__.py) Hey anybody help me asap

[@NicholasRenotte]

Heya @Starboy RaM, can you try going through this: kzbin.info/www/bejne/movLkKOpbcx-ndU

[@ammarsaper2650]

Bro can you help me ????

[@NicholasRenotte]

Yo, what's up?

[@ammarsaper2650]

My final project is in sign language translate (translate it to written language ) so until now I don't know-how. A collection of my dataset.

[@NicholasRenotte]

@@ammarsaper2650 check this out, will show you how to do it step by step kzbin.info/www/bejne/pnW7lZ-OocZ2mdE

[@ammarsaper2650]

I ‏saw all your videos it's very important I love it. But the problem is how can we deal with the long words, after translating it from sign language.

[@NicholasRenotte]

@@ammarsaper2650 you can also use action detection! I'm still working on a video for it but there's also a pre-trained Tiny Video Net model available on Tensorflow Hub that can be used.

[@isaackay5887]

First off, these signs are not accurate ASL signs at all. Second, ASL is a full-blown language - it's so much more complex than "single-shot" hand poses...this isn't even scratching the surface on what needs to be done. Third, you should've used video footage from Deaf KZbinrs - they'll always be more accurate than you or me. Came here for reference because I love your videos, but kinda left here disappointed seeing that there wasn't any prep work done on sign language(s) at all. (Btw, I love your enthusiastic spirit and your teaching style. I'm in no way trying to put down what you've built here - it's impressive and certainly beyond my capabilities at the moment to produce this from scratch like you did - but I thought it was worth bringing to your attention if you're wanting to serve the Deaf community with this...)

[@NicholasRenotte]

Heya @Isaac, definitely, I think it needs a helluva lot more work before this is to the level where it's readily deployable and useful for the ASL community! That being said, it's a first step in the intended direction of where I'm hoping to take it. The core thing as you said, is that this model isn't currently using multiple frames but rather just detecting based on a single image. The plan is to take this use case and extend it out to leverage an action detection model, like tiny video net, and train it on significantly more classes. No offence taken, I welcome feedback and I'm always looking to improve, both myself and the tech! Keep at it ✌️