Powered by TensorFlow: Air Cognizer predicts air quality with machine learning

Рет қаралды 83,012

4 жыл бұрын

Machine Learning is helping to solve challenging, real-world problems around the world. Poor air quality is a major obstacle that many face. See how Air Cognizer is using Machine Learning, powered by TensorFlow Lite, to help predict air quality using user images and eliminating the need for expensive, portable air quality meters.
Check out the blog → goo.gle/2DWHcQ5
Check out the website → goo.gle/2CqagyG
Add the Powered by TensorFlow playlist → goo.gle/Powered-by-TensorFlow
Subscribe to TensorFlow → goo.gle/TensorFlow

Пікірлер: 206

@MB-hz7wm 3 жыл бұрын

I’m currently learning TensorFlow and appreciate you posting a real world application example for bigger picture comprehension for those new to the platform!

@viniciussantana8737 4 жыл бұрын

This is a Very Clever use of ML. CONGRATs from Brazil!

@chaitanyamogal808 4 жыл бұрын

Indian needs engineering's like you .......all the best for your future 😀

@aruneshwarv1051 4 жыл бұрын

I like this channel a lot,, just got started with tensorflow ,, it's amazing,,,

@RandomGuy-hi2jm 4 жыл бұрын

The project is basic but the way u put thay was amazing, well done guys.

@steved.1091 3 жыл бұрын

I agree. It's the application that blows me!

@redsmile3663 3 жыл бұрын

@@steved.1091 This application is not usable in a real life scenario. They have trained on one city only and its highly polluted city. If they use a cleaner city dataset with it to train, i highly doubt the accuracy will be good enough. You can not get AQI just using images. There are too much variable involved that images can't captur. How will u differentiate fog from smog.

@steved.1091 3 жыл бұрын

@@redsmile3663 So, should I not compliment them on their achievement? 🤷

@redsmile3663 3 жыл бұрын

@@steved.1091 I am just saying that the idea of predicting aqi using just images should not blow u away bcz there are some other parameters associated to it that can not be gathered using images

@shelendravashishtha9233 3 жыл бұрын

@@redsmile3663 I think it doesn't matter if there are images with every kind of air indexes then it doesn't matter if these pictures were taken in one city or so many cities....they have already included so many locations so there is no need of various cities in my opinion at least

@Madhuram_Qualityoflife 3 жыл бұрын

Well done team! Keep working on models that make this world a better place.

@josephjose7373 3 жыл бұрын

Please read the blog in the link before asking questions in the comment section. They have done some really amazing work.

@samarthsaxena2703 4 жыл бұрын

This app is really helpful...thanks to these amazing people😊

@Pooja_gujral 4 жыл бұрын

Amazing innovation by this duo💯 tried and tested their app and it works well. Kudos to brilliant minds!

@jyothiswaroopsantu 4 жыл бұрын

Great! It's really good. Hope you develop this more towards other cities too and make this accessible to everyone. All the best !

@dg49626 3 жыл бұрын

A great use of tensorflow. Amazing work team!

@nikitanaik2210 4 жыл бұрын

This is amazing. Looking forward to something that would fix the air quality problem.

@Racers_Club 4 жыл бұрын

Really appreciable... A great team work

@hnkulkarni 3 жыл бұрын

Very good application!!! 👍🏾👍🏾

@prashantsharmastunning 4 жыл бұрын

simplistic yet effective thats the best solution.

@vivekanandr8120 3 жыл бұрын

Let me install it right away! 👏👏👌 great job guys!

@raghavbali3671 Жыл бұрын

Bahot hi umda!! Kudos Prerna & Tanmay.

@shashwatpandey5165 4 жыл бұрын

Motivation level: This video :)

@s.mswapna3464 3 жыл бұрын

Beautiful and never ever I thought this idea exists and u made it🙏👏🤝♥️

@mithsaradasanayake3211 2 жыл бұрын

Thank you. Awesome work

@praveenkumarverma3767 3 жыл бұрын

Very nice, well done guys..

@ritiksahu5310 4 жыл бұрын

People are doing so many great things . Me still figuring out how to import tensorflow #error_error_error 😩😩😭😭😭

@kishankalariya9086 3 жыл бұрын

😂😂

@qmobiletab4773 3 жыл бұрын

Don't lose hope, every one goes through this timeline.

@ritiksahu5310 3 жыл бұрын

@@qmobiletab4773 🙂

@JaspreetSingh-fo2qe 3 жыл бұрын

Check if the processor on ur PC has AVX support. This caused a lot of trouble for us for days. If AVX support isn't there then you can work on google collab.

@harsh9558 3 жыл бұрын

I dont even know what tensor flow is🤣🤣

@parulg014 4 жыл бұрын

Remember to classify fog and pollution!!

@mwanikimwaniki6801 3 жыл бұрын

That will be tough

@parulg014 3 жыл бұрын

@@mwanikimwaniki6801 I would say it would be tricky :)

@MereDaddyJi 3 жыл бұрын

thats easy

@parulg014 3 жыл бұрын

@@MereDaddyJi how?

@Bodydrop93 3 жыл бұрын

Yes indeed it is tricky with just an image. Maybe is possible to extend the input data with some weather API data (sth like pressure, humidity,...)

@ParthPatel-fw5zt 3 жыл бұрын

Nice to hear that someone tried to predict the AQI index.

@tamojitdas7643 3 жыл бұрын

A really good initiative

@made2automate 4 жыл бұрын

Just Amazing!

@DrLouellLSala 4 жыл бұрын

Great job with machine learning. I'm using tensorflow in some of my projects as well and created ML apps in Healthcare. I'm interested in using this for Covid19 datasets though

@MereDaddyJi 3 жыл бұрын

take me in team

@andreasjanzen8823 3 жыл бұрын

Cool project! Well done!

@The369niraj 4 жыл бұрын

Not sure if an image of an sky is really a good way of predicting AQI. Instead they should just give binary classification of safe & not safe.. This will be like quantizing the index range into 2 bins and will be less prone to errors

@RandomGuy-hi2jm 3 жыл бұрын

Using regression (predicting AQI values) makes time series prediction (forecasting) more accurate.

@mgtgaming55 4 жыл бұрын

Amazing discovery...all u need is to think and blieve in urself. Nothing is impossible

@adwaitnaravane5285 3 жыл бұрын

Except when your GPU starts acting up.

@kashishsharma6809 3 жыл бұрын

Try it, works on cat image also😂

@harsh9558 3 жыл бұрын

@@kashishsharma6809 wait what?

@rahul_ji123 3 жыл бұрын

congratulations team fantastic job

@kavibharathi2931 3 жыл бұрын

Great work

@karanpahwa7866 4 жыл бұрын

Amazing work 👌👌🙌🙌

@sudhanshuupadhyay458 4 жыл бұрын

Really good concept. 🔥

@mayankdandhare3342 3 жыл бұрын

Kudos for your efforts ✌️

@bijeshmohan 3 жыл бұрын

This is amazing!

@martharajkumar7995 3 жыл бұрын

Amazing model

@lecturesoniesgatepreparati8554 3 жыл бұрын

Nice project guys..

@Mohanraj-li9ff 3 жыл бұрын

Its a great idea... Congrats

@DrSirmacek 4 жыл бұрын

What a nice idea!

@alanlee4260 3 жыл бұрын

woow it inspires me a lot in this major

@AjayYadav-xi9sj 3 жыл бұрын

Wow it's really amazing 🙌🙌

@bhuvaneshs.k638 3 жыл бұрын

👍👍😍 awesome work

@prateekgaur1172 4 жыл бұрын

Just amazing guys

@munindarahir9934 3 жыл бұрын

Great job guys keep going.Best wished ahead

@dolendeori3912 3 жыл бұрын

Its amaizing good keep it up 💓💓

@user-sk3to3nl9h 3 жыл бұрын

You must save data in cloud... This wil help increase accuracy by letting ppl but some note if they want and also by gathering info for that day in the city and build algo or other model to make sure users in that city get almost the same results with push notification if there is an update Also predicting the air cond for tomorrow using the weather and if its rainy or not... Etc

@shadabahmed4547 3 жыл бұрын

Good work

@iraiarul25 3 жыл бұрын

Great guys👏👏👏👏👏

@sanjaybhatikar 3 жыл бұрын

Good job 👏

@aulalivre5481 2 жыл бұрын

I think this best way to use tecnologies.

@architjain5108 4 жыл бұрын

What accuracy you get? N is dataset open source?

@skhapijulhossen6499 3 жыл бұрын

So Inspiring

@mossaimelyse 3 жыл бұрын

Pretty bright wow !

@user-ye7vg6fh1b 5 ай бұрын

really helped us a lot!!!! Lyk we r few AI students!!!!!! presentation its superbb!!!!

@Viralvlogvideos 3 жыл бұрын

I want to know the accuracy of this model with real air pollution testing device

@shreehari2589 3 жыл бұрын

Awesome guys

@samshaikh278 4 жыл бұрын

Does it give current AQI values or predict ?

@indrajitmukherjee5964 4 жыл бұрын

How to remove noise from the images? And what's the accuracy leavel of the actual payload?

@gundakoushik7840 3 жыл бұрын

Great thing use to people

@ayeshanoor1027 3 жыл бұрын

well done

@miratik8507 2 жыл бұрын

Very impressive

@tr911tr 11 ай бұрын

Awesome 👍👍

@kenyup5424 3 жыл бұрын

This case of implementation is useful for China as well, I like the positive vibes in this video,it inspired me

@ameyaupalanchi5944 3 жыл бұрын

You can get the data from website also

@alikarani1014 4 жыл бұрын

Amazing

@AshishChauhanYoungy 3 жыл бұрын

This is really good stuff, and TF Lite makes it even cooler! I wonder if code behind this project is open source !!

@icaruseden1783 4 жыл бұрын

Cool .... I believe the model training phase must've been the most tedious one.

@gunaseelanchinniah1639 4 жыл бұрын

Good initiative, There are so many variables when it comes to a reasonable prediction. I suggest having at least 50k images instead of 5k. What was the training and validation accuracy? thanks and are the 5k images on public domain?

@gokuladethyat1466 3 жыл бұрын

i support your suggestion too.

@armaangupta9986 3 жыл бұрын

Dude this isnt your mnist dataset they are collecting the data themselves and the also have their lives and collecting 5k images by just some students is really hard like they are students they have to attend classes too and despite all these they collected this many images if it was a big company such as google then it would have been easy! And as for the training accuracy i dont think they wanna reveal it so let it be their style and yeah the 5k images wont be publically available cause obviously data is the fuel and these data are really precious so why will they give them off

@ujjawalx7460 3 жыл бұрын

Bump

@harsh9558 3 жыл бұрын

Not an ML expert but read that modern AI will do work will less data🤔🤔 If it's TRUE it will make work very easy

@armaangupta9986 3 жыл бұрын

@@harsh9558 ohh nooo infact deep learning ais need even more data as compared to normal machine learning algorithms

@deepakpratap3792 3 жыл бұрын

Wow...cool project...AI for humanity

@maheshvangala8472 4 жыл бұрын

Good initiative

@pravinborate1500 3 жыл бұрын

Where I can get dataset for same

@avijay70s 3 жыл бұрын

Awesome

@jotaro8694 3 жыл бұрын

Great

@kashafkhan8535 2 жыл бұрын

Nice

@sandyz1000 3 жыл бұрын

I have lot of genuine queries. Let me begin 1. Is the quality of image will affect the prediction, considering the prediction has been made based on how clear the image taken. A ideal image will be one that has least amount of noise in an image taken from the camera sensor and without applying any pre-processing on it. 2. How are you defining the Mean opinion score of the AQ, since all those images are manually labelled. Based on my understanding on the subject, image features is just an extra dimensional features that taken in conjuction with meteorological data help you to determine the score, but even in that case image features will have low eigen value for predicting the score.

@swapnilraj3653 3 жыл бұрын

Will it classify the image correctly if taken from a dirty lens

@akshitbansal352 3 жыл бұрын

How are the results affected by gamma correction on the image. I am not sure if an image is a right choice for determining air quality. Anyways, well done guys.

@DhirajPatra 3 жыл бұрын

Nice where is the code and architecture can I see?

@nandakumart2331 3 жыл бұрын

I like that way 👌

@amanrastogi5184 3 жыл бұрын

Your attempt is very good but i would love to know that how an image can measure the levels of Carbon Monoxide, Sulphur dioxide, Nitrogen dioxide etc. which are the sole contributors while measuring AQI. Thank you.

@rohansingh1057 3 жыл бұрын

Air quality has other factors. PM 2.5 and PM 10 refers to the size of pollutants. This can be measured by observing a ray of light in the air. Not sure if the camera can sense up to that accuracy though. But ML is very good at identifying patterns that humans cannot.

@amanrastogi5184 3 жыл бұрын

@@rohansingh1057 so what about the toxic gases Sir ji. Because your motive is not to check only the air particle size but to check the quality. The mucous of you nose and the hair in it can easily deal with big size particles but not with tye gases anyway. If the used algorithm is able to check the level of toxic gases and then can measure AQI then its pretty helpful otherwise you don't need it at all.

@hiramcoriarodriguez1252 4 жыл бұрын

1:23 it's image segmentation not image classification. After image segmentation then they do image classification

@RandomGuy-hi2jm 4 жыл бұрын

It can be said classifying also since they are easily linearly separable.

@karandua6564 4 жыл бұрын

It's classification. It's not segmenting the image and just telling if the image has sky or not.

@hiramcoriarodriguez1252 4 жыл бұрын

@@karandua6564 at the end, as i and you said, it's classification but by definition they use first image segmentation.

@karandua6564 4 жыл бұрын

@@hiramcoriarodriguez1252 No, not really. The image they used was a bit misleading. It's not necessary to "segment" the "sky" out of the image in order to check if sky exists in it or not. It's just a yes or no question - a binary classification problem.

@chirag_samal 3 жыл бұрын

Is there any GitHub implementation??

@Vision267 3 жыл бұрын

How do you commercialize this app

@vaibhavkulkarni5601 4 жыл бұрын

What are the parameters for meteorological model ? The app must be using an api (to get these parameters based on geo locations)which gives them this data to process on. Well that would be my approach on it Any suggestions about it ?

@redsmile3663 3 жыл бұрын

They are just using image as a parameter. I think its not accurate. Bcz air quality depends on the constituents in the air and u cant possibly get that info from images. Also, there will be problems like fog vs smog. I think this is the reason they did not share their accuracy.

@pravinborate1500 3 жыл бұрын

Where I can get dataset for same?

@usmanyousaaf 8 ай бұрын

can you please provide the image dataset

@sushantrauthan5704 4 жыл бұрын

Great job folks!!! I 'm also a resident of DELHI and am fed up with the current air pollution crisis that DELHI faces , so a while back i started to learn machine learning to create a ML model for air quality prediction although I segued into creating a ML model for predicting climate which in turn helps while setting up a solar panel , i can comfortably say that currently i have an intermediate level in ML and and have recently started learning DL , please ping me if you guys need some extra skilled help , also i have a lingering question in my head regarding your project which is:- a photo depends on many things such as ISO , shutter rate , the type of software the camera uses , the quality of camera and so different cameras or different camera settings might end up producing different images of the same ambient atmosphere , so dosen't your model gets confused or produces results which are not actually true? Thanks Warm Regards

@fahimsanghariyat162 3 жыл бұрын

Yup .......

@joshua_v_thomas 3 жыл бұрын

How will it differentiate pollution and fog? Just out of curiosity 😉

@chanduvamsi1841 3 жыл бұрын

Good one but don't you think images of a single city is enough to classify all the ranges all around the world? I think 5k images are pretty slim and atleast use 50k images to improve your accuracy. And also the image quality and hue and everything else changes from device to device. Please look into it. Anyway good work, guys. Keep learning ❤

@arnavdas3139 4 жыл бұрын

Wuuu cool

@savalalingeshreddy6750 2 жыл бұрын

Very nice project I am interested in Machine Learning I want to do this project on my own how to start? i am not getting any idea

@ANKITKUMAR-tn8lt 3 жыл бұрын

Can this app identify btw the fog and smog??.....bcz in pictures both look the same. And my second question is how much depends this app on the quality of phone?