How to Make an Image Classifier - Intro to Deep Learning #6

[@mayurbhor2231]

I come to this channel only to get depressed and know that there is still lot to learn

[@Mohammed-rx6ok]

+1

[@raphaelrodriguez1856]

Lmao it’s the complete opposite for me! I get so hyped!!!

[@praful0203]

I became a great fan of you Siraj and I started exploring Machine learning and Deep learning with your videos. Though in earlier stage I'm not understanding everything you explain but with full dedication and smart work, I'll also become a Programming Wizard soon. Just keep sharing your knowledge, I love the way you teach by merging the content with your raps. Thank you and God bless you

[@CalvinEchols]

I think this is the first comment I have left on KZbin, but I had to leave this to tell you that this video is the best description of how a CNN works that I have seen. Love how you break down how what the individual layers are doing and how the calculations work at a high level. Thanks for the awesome video!

[@10078698]

I'm sure this is the first comment I've ever posted and I agree! your explanation of each layer of the CNN has been the clearest, most bs free one I've come across! thank you! that said, it would be even better if you could fit in a description of how the final FC layer works or what it does, and I think it would be perfect!

[@Treegrower]

He didn't really cover how CNN's expand and contract the dimensionality of the input, which I think is a really important concept that he glossed over. Other than that it was good though.

[@SirajRaval]

im honored, thanks!

[@aashray18521]

Do go through this link in description adeshpande3.github.io/adeshpande3.github.io/A-Beginner's-Guide-To-Understanding-Convolutional-Neural-Networks/

[@maniksony]

Why is the matter written on your link is same as what Siraj said??

[@cupofkoa]

"we're going to be building a CNN from scratch" - then uses Keras

[@The_midnight_motivator]

If want it from scratch see sentdex videos

[@huynhsang9835]

NO 'load_data' in 'parser'

[@sivakrishna1682]

Found no relation between the time you took and you explanation. Simply amazing !!!

[@StevenSmith68828]

Why are people so mad about it not being beginner friendly, machine learning is a very complex subject. In fact even this is easy in comparison to what each one of those functions are doing. He tried his best, I personally don't see how I might simplify this further.

[@garettgazay9957]

You really are the King of Teachers, Siraj. Thanks for this, you always explain ML concepts so well, and the videos are never boring.

[@SirajRaval]

thanks Garett i'll keep improving

[@adowaconan]

Thank you very much for make so many informative and practical videos!! I enjoy re-watching most of them many times to get a hang of them. I am actually from a background of auditory and memory, thus, always wish to implement what I learn here to my field of research.

[@SirajRaval]

thanks! anytime

[@Noone-wl7ld]

so sad... This vdo seems informative but presented steps are too advance to follow (many guidance missing). it is not helpful to beginner like me. :(

[@RajeshPachaikani]

Same feeling bro

[@444whoislex]

That’s why it is episode 6

[@sciencesbasicnatural9042]

Agreed

[@code-grammardude5974]

If you learn about neural networks then convolutional neural networks, it will increase your understanding very much. It may seem daunting but it isn't too bad after reading/watching a bunch of explanations. :)

[@suryanshpatel604]

You need to watch the earlier 5 videos.

[@victorminero]

Keep rapping! Keep teaching the way you want. Is for that spontaneity that 65,000+ we are following you. By the way, great lesson. Greetings from Mexico City!

[@SirajRaval]

You got it Victor, thanks! Viva Mexico!

[@satyamdwivedi8135]

aree bolo is chutiye ko ya to pdhaye ya to nacche

[@UsmanAhmed-sq9bl]

Top quality Siraj. You and your videos improving day by day. Excellent work.

[@SirajRaval]

thanks for noticing Usman :)

[@Colab-Film]

I love your videos! You do such a great job in explaining such complex things and making knowledge accessible! Thx!

[@chrisoffersen]

MY GOD, THANK YOU.

[@SirajRaval]

awesome lol thx

[@elyay7203]

Great video, it works! For anyone trying to use his code, you can either downgrade keras to v2.0 or make these changes to the 3rd step: sample_per_epoch ==> steps_per_epoch, nb_epoch ==> epochs and nb_val_samples ==> validation_steps it took me a long time to fix my problem, but i hope you find this as soon as possible

[@apoorvsxna]

Thanks! One more thing- from parser import load_data shows "cannnot import load_data from parser". Any idea how to fix that?

[@akfaisel]

Your English is so beautiful!

[@prathaprpm]

you are surely KING of teacher

[@AkashMishra23]

All your Intros are Cool af

[@SirajRaval]

thanks Akash

[@AlizerLeHaxor]

Probably the most beginner-friendly tutorial on Deep Learning I ever seen.

[@vincentsahler5311]

Very good video I've learned so many think ! Thank you and please continue !

[@Hatim14031879]

Man I really like your videos, Big up from Germany !

[@SirajRaval]

danke!

[@0waco0]

Your videos are great! please continue to make them.

[@SirajRaval]

will do thanks Hassan!

[@bizharshit]

code is wrong image is not defined img = image.load_img('dataset\test_set\cats\cat.4014.jpg', target_size = (224, 224))

[@maheshkoheda9194]

Just get going bro!love your appproach for topics.Thanks great one.

[@mesinjah]

bro i love your style of teaching. Chur!

[@cehasli]

The coolest teacher of ML rsrs

[@SirajRaval]

thanks!

[@nikola3013]

Thank you for the explanation, very easy to understand. :)

[@averyc7492]

Wow, this video is the best thing ever. Coming from someone with 0 python experience.

[@SirajRaval]

awesome!

[@ABOSEDEPETER]

You full of life....love the teaching method

[@zubairuddinkhan5075]

this guy is so underRated :(

[@shahzadaliii]

from parser import load_data gives following error: ImportError: cannot import name 'load_data' from 'parser' (unknown location) how to resolve this.

[@TheMujumuju]

I can't believe he actually sang in a CNN tutorial :D I want to be as high as him!

[@iamdurgeshk]

Please don't be boring like others by not doing rap. It's your way of explaining things then let it be. Great video. I have started to understand things little bit. :D I will start working on it practically soon and will drop few questions which I will have for sure. :D

[@SirajRaval]

thanks will do Durgesh

[@angelachikaebirim1043]

love the rap

[@abirmohammed17]

Can anyone tell me how to install the "parser " package? i can not seems to compile the line : "from parser import load_data"

[@elvisun]

B M Abir try using keras' built in image preprocessing and loading library, it's really easy to use

[@abbadabbajabba5966]

bro use this..it works for 101% from keras.datasets import imdb train_data=imdb.load_data('yourdatapath')

[@10yearsofpython91]

@@abbadabbajabba5966 Thank you

[@prestonwetta2510]

the rap made me subscribe coz i like how it describe almost the whole scribe!

[@rudhisundar]

This is going too fast and is really convoluted!

[@donkeroo1]

Rudhisundar Beura because he’s full of Shit

[@andyeeckhaut]

Noob question: I get an error on this line model.add(Convolution2D(32,3,3 input_shape=(img_width,img_height,3))) and I seem to overlook what I did wrong. Any help? :-/

[@stumblinzz]

I really enjoy the series, but I think when I initially watched this video upon its release I struggled with the core content at first because its easy to get wrapped up in jupyter or get stuck installing something, which creates a lot of additional friction when I need to learn the core concept.

[@ArShaykh]

I hope you will explain in the live stream how the weights of the filters are being learned. I know that it is done through backpropagation but i just don't understand how to update the weights since filter connections are so different from usual connections between neurons in other NNs.

[@LarlemMagic]

Those images of the feature map really helped me understand.

[@21RandomMan]

Tensorflow and Keras are very "from scratch" tools in the industry. These tools are made by tons of people collaborating and putting it together. If you don't understand what Keras abstracts, go read up on what a CNN is and what activation functions are etc. If you DO understand when to use say... linear regression or gradient descent, tools like Keras becomes really helpful. He said "Make our own Image Classifier" not, development a deep learning framework.

[@rahulkrishnan529]

At 4:09 what did you mean by feature mapping.

[@543soldier]

Hey Siraj. In my "collect data" cell, I get an error saying 'NameError: name 'load_data' is not defined' . What program is this load_data command coming from? Do you think you could point me in the right direction?

[@swatipetkar6852]

yup.. thats wrong there. You need to use keras ImageDataGenerator and flow_from_directory to get this working

[@vedantmhatre7235]

Your code is not working

[@yuvanbajjurla9813]

Siraj could you please make a video on image processing using tensorflow and python

[@cbxsn]

are you going to do a video on face recognition? e.g. facebook's tagging feature

[@SirajRaval]

yes soon

[@dr.dietrichdavidstein8425]

i know you get a lot of requests, but could you please do a longer video (like your 1h sessions) where you describe neural nets that can generate audio/music? i've seen a lot of nn midi generators - but only 1 or 2 that work with raw audio. Something like this: github.com/MattVitelli/GRUV love your work - keep it coming!

[@gowthamanv8740]

Which database is suitable for deep learning for large amount of data?

[@khalidakram557]

For people who don't understand his way of teaching like me and want simple and crisp way to teaching, follow deeplizard on youtube. She is the best KZbinr for me so far.

[@TidalMaker]

I love your videos, especially the cringe!

[@SirajRaval]

lol thanks!

[@amomasi9909]

One side thing I appreciated from this tute (as a beginner) is the sheer amount of space and computation power required to do all this. :'(

[@marcograziano7308]

Very well done! How about a follow up along the lines of your original and great "build a Tensorflow classifier in 5 minutes"? Inception transfer learning in Keras perhaps?

[@SirajRaval]

thanks! Live stream coming wed will be related

[@geethzzzp4384]

Can you do a video on image classification using Extreme Learning Machine..???

[@canalper4866]

Would it be better if I learn Tensorflow before jumping into something like Keras ?

[@redph0nix]

Really important question. I am new to machine learning and I am confused about detecting multiple objects in a image. I am working with tensorflow, but I am confused on if i can detect multiple objects in an image with JUST tensorflow, or i need a library like openCV to help detect objects and then classify them with tensorflow? If I do not need openCV, how would i do this in just tensorflow. Any help would be great! Thanks in advanced!

[@ArshiaBiswas]

woww this is aweosome Can u tell me if i can import the dataset in VS code too? if im going to run this whole program there?

[@wouterrobers224]

When i try model.predict(img) I get Failed to find data adapter that can handle input: ,

[@raghurrai]

I just had one doubt. Suppose that we have a filter to detect the curve as shown in the video but the curve is aligned in some direction. How would it still recognize it? Or is it even supposed to? In some cases, like identifying an object, the filter should be able to identify the object regardless of it's orientation. While in other cases like identifying numbers, it shouldn't cause it may confuse a 6 with a 9.

[@johndoucette3687]

I wish you would have left a link to the code for each video. Not too sure I am ready for challenges but would like to see the 40 lines (or however many were used in video) of code from the current video.

[@aakashaggarwal007]

can someone explain the load_data function that was used in video

[@shayeransari9495]

see some other tutorial lot of are available, he makes worst tutorial, explaining nothing. just bla bla bla

[@kingmemo81]

you are inspiring!!! dont stop!

[@SirajRaval]

thanks!!

[@user-qx4bg1jo3y]

ImportError: cannot import name 'load_data' from 'parser' (unknown location)

[@venytonizidoro]

im getting the same thing

[@ShaikAsad]

blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html

[@virtiparekh1394]

after clicking image from camera i want to classify the image by pressing button is it possible with this?

[@MBA2804]

Can you add the load_data function or parser file to the description pls?

[@uttamdwivedi7709]

hey siraj, great work :) but I can't import load_data from parser :( :( help help !!!!! >>> from parser import load_data Traceback (most recent call last): File "", line 1, in ImportError: cannot import name load_data

[@666aron]

I have the same problem, still couldn't figure it out.

[@exec9292]

check the path

[@ashwanijha7643]

Hey! Did you get a solution to this?

[@bishnupriyapradhan7768]

Hey, Could you figure out a solution?

[@PranayAnkit]

parser module does not have any function named load_data. I think he has defined a custom module and put load_data function inside it. You can have a look here to read images: github.com/rdcolema/keras-image-classification/blob/master/cats_n_dogs.ipynb

[@Supergecko8]

What about classifying 4 images to identify the course of an object?

[@SirajRaval]

github.com/tensorflow/skflow/issues/113

[@shreerangaraju1013]

I was wondering if you are gonna do a video on apache spark and mllib.

[@SirajRaval]

hmm will consider

[@ashabb_official]

ImportError: cannot import name 'load_data' from 'parser' (unknown location)

[@smit4308]

+1

[@AnimeSongsNightcore]

How did you fix it ?

[@ashabb_official]

@@AnimeSongsNightcore i didn't fix yet

[@daasdaham6964]

man was that intro DEEP !

[@Someshsunariwal]

What is the best way to make a alphabate classifier.? What is better approch using tensor flow only or keras only?

[@madhukumars458]

I didn't get , Where did we label the image? or please explain the training data format

[@logicaldistraction]

"from parser import load_data" seems to be a custom python module!? could you provide the code for it somewhere?

[@phuccoiinkorea3341]

you're right. i fucked up this

[@abbadabbajabba5966]

bro use this..it works for 101% from keras.datasets import imdb train_data=imdb.load_data('yourdatapath')

[@pranavbaradkar5763]

@@phuccoiinkorea3341 same man. i fucke dup this and nothing getting out of this.

[@monivibes1]

Hello Siraj, I want to get your opinion on a project that i'm starting: It is a deep learning and i feel that i need some advice to know in which direction i can go. Let's start: I want to use deep learning to build a model that recognizes many characteristic in an image. To be precise, i want to create a model that allows me to recognize in a photo: .if the person is a MALE/Female. .what type of clothing is he/she wearing .what are the colors of the clothes he/she is wearing. What do you think of the complexity of the problem ? I really need some guidance in order to start thinking of the possible techniques that could help me go deeper in this project. Do you advise me to start thinking first of a model that can olny distinguish male/femal (and ofc it needs to detect them first)?Or do i need to think of the subject from another perespective? What are the topics that i need to look for? Thanks in advance.

[@computerscience5242]

what is parser????

[@iancarrasco5430]

Hey Siraj, how would you use the saved adjusted weights in a CNN similar to the one you made?

[@iancarrasco5430]

Thanks!

[@tt700joe]

So has anyone figured out the from parser import load_data mystery? I'm yet to find the package and can't continue if I don't have it! P.s. if anyone has any idea on how to create one of our own we can do that!

[@tt700joe]

Ended up solving this myself, if you're interested in seeing another tutorial, particularly the one that had the explanation for how to load custom data sets here is a fantastic one right off the keras blog! It's almost identical to Siraj's blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html

[@divikshrivastava6476]

@@tt700joe thanks a lot

[@rejusubash5297]

I have a huge seminar on image recognition tomorrow in college and i am watching this noob rapping about it

[@rancegarrett3696]

First time seeing you! Awesome! It's just that i'm still in high school and i'm a noob-ish programmer, which explains why i lack full understanding. ":/ I'll get there soon...

[@SirajRaval]

Hey Rance! Welcome to the most important community in the world. Excited to see you progress in your journey

[@carolinetv1968]

It is really very useful!!!! thank you!

[@SuperReddevil23]

Can you please do a video for image classification using K nearest neighbours

[@prajankya]

@2:19 King of Teachers !! Swagg.. nice..

[@SirajRaval]

thanks! lol

[@numoru]

You say 32 by 32 for the first convolution2D added to the model , yet what is the difference if we used 64 by 64. Would we gain resolution in data and accuracy in exchange for more processing time; or is this difference negligible and the results nearly equivalent?

[@jonathanclark6489]

6 years ago?? sheeeesh i was expecting 6 months lol

[@simonkerr9255]

Awesome videos Siraj. I'm new to this, but have a classification problem in mind. One question - Would a TensorFlow image classifier be rotation invariant? In the Cat/Dog example, would it still identify cats/dogs of all rotations? What about images of circular objects like, say, coins. All the same size, photographed flat, but arbitrarily rotated, with different markings distinguishing the classes?

[@THESocialJusticeWarrior]

what about using a classifier to pic what direction the animal is facing THEN have it pic a cat/dog classifier that is trained on photos taken at that angle? How would you do that? Wouldn't that be much more accurate?

[@rodrigomataibanez8859]

THESocialJusticeWarrior You can basically use the same code, just use the training data of your dogs / cats in the angle required. The way this works is by the convolutions, or the data distribution in matrices. It really doesn't 'know' if it's a cat or dog, it just gives the closest approach. That's why there's an accuracy, and also the reason to avoid overfitting.

[@look3248]

Loved that intro

[@SirajRaval]

thanks Luke

[@bizharshit]

image is not defined !!!

[@vikasdalal5159]

I understand the concept how backpropagation help to optimize the weights in fully connected network. But how backpropagation help to optimize the weights in filter of Convolutional Neural Network ? Can somebody explain ? Thank you.

[@naveenkumarm8828]

ImportError Traceback (most recent call last) in () 2 import numpy as np 3 import parser ----> 4 from parser import load_data ImportError: cannot import name 'load_data' --------------------------------------------------------------------------- NOTE: If your import is failing due to a missing package, you can manually install dependencies using either !pip or !apt. To view examples of installing some common dependencies, click the "Open Examples" button below.

[@profi45]

I got the same error. What did you do to solve the problem? Thank you.

[@thexxmaster]

Youre a god and I love you - thanks a bunch

[@SebastianRaschka]

Interesing video! just as a little bit of feedbaack: you are using a weird terminology for derivatives :). E.g., at min 6:50; by using partial \delta = \partial J / partial s you may end up confusing a lot of folks who are familiar with the delta rule.

[@SirajRaval]

thanks for the clarification! Love your book. I'll annotate it

[@SebastianRaschka]

Cool! I've been watching through your other videos this morning, and whoa ... never seen such an enthusiastic and energetic educator! :) Thanks for putting these together and sharing them, they are pretty entertaining and definitely brightening up one's mood on a cloudy Sunday :)

[@grantgre]

Will enjoy your videos very creative. Where are you been come back it’s been a couple years I think. Come on back baby.

[@mouniswar7]

Is it possible to train Deep Learning Models on Gt840M GPU?

I love your channel !!

[@SirajRaval]

thanks!

[@sheriffola]

CNN just got stuffed in my head! Great video, clear explanation. Siraj should really consider making a book on Machine Learning.

[@SirajRaval]

awesome, thanks! I want to, i need to find the time. hmm. probably this summer.

[@sheriffola]

Ok.

[@valken666]

+Siraj Raval If you add a way to add filenames to tfrecord you'll be my God!

[@priyanksisodia]

Intro is motivating

[@alexalexfff]

why last dense layer only one neuron size? It 0 for cat and 1 for dog?

[@viniciustozzi8786]

Man your videos are very very good, and the rap is amazing! But I am confused about something, what exactly relu does it? Just change linear values to non linear values?

[@LapajHaver]

Hey Siraj, why do you use a sigmoid in the output layer? is it because it's a binary classification?

[@SirajRaval]

exactly

[@samuelvarghese6881]

+Siraj Ravel hey would you do more keras tutorials as tensorflow is going to include keras in the coming version

[@SirajRaval]

sure thing

[@faatemehch96]

can i use this method for detecting covid-19 by X-ray images?