Build a Deep CNN Image Classifier with ANY Images

[@Jey187]

For those that are frustrated by the first line of code not working, just remove the "tensorflow-gpu" part and run the code. You may want to skip the GPU setup part cause it won't work anyway. The algorithm will work, except slower (mine is 656 ms which takes me less than 1 minute to go through the entire epoch of 20)

[@bytes134]

hey, do i only remove tensorflow-gpu or the tensorflow too? because tensorflow is not working on jupyter notebook

[@Jey187]

@@bytes134 I think tensorflow should be working. You only have to remove the GPU one

[@nicholashendricks9762]

You have to know how to get the stack/drivers setup for your GPU disabling GPU support is a horrible idea seeing as that's what all modern modes are literally optimized to run on. Wether you have an AMD card and need to setup rocm or just simple nvida and cuda setup.

[@elbayar5188]

thank you

[@pradyumnanr3680]

tensorflow-gpu needn't be installed for TensorFlow 2.x or higher. The GPU support is built into the main tensorflow package. If your system has has a compatible GPU and the necessary drivers (NVIDIA GPU driver, CUDA Toolkit, cuDNN Library) installed, TensorFlow will automatically detect and use the GPU

[@venomlovekitties]

As a non coder person I instantly subscribed because of the simplicity you showed by your teaching skills. Thanks man, love to see more content from you.

[@CantPickTheNameIwant]

as a CV engineer, I instantly hit the dislike button under this video

[@mihai3678]

@@CantPickTheNameIwant that’s what I wanted to say 😂😂big source of misinformations on this channel, specifically in this video

[@daryladhityahenry]

@@mihai3678 Can you tell which one is misinformation and how should it be? So I can know which one that I should look for... THank you....

[@akikabanu5885]

@@mihai3678 how come? do you think you could explain?

[@muhammaduseram9405]

@@CantPickTheNameIwant at least you should clear your point if you said it

[@marti-nz]

This tutorial is amazing, not only are instructions easy to follow but sufficient explanation is provided so I know why each line of code was added. Great Job!

[@AlvaroPerezLuque]

Hi Nicholas, the tutorial is fantastic. There is a small bug in the code. The data object is always shuffling the data so there is no difference between training and validation data. To fix this bug it is necessary to change these lines: 1. Edit the line data = tf.keras.utils.image_dataset_from_directory('data'). It should be: data = tf.keras.utils.image_dataset_from_directory('data', shuffle=False,) 2. After this line add the following: data = data.shuffle(1000, seed=100, reshuffle_each_iteration=False) Without this correction the val_accuracy will always be 1 as there is the same data. (The problem arises when calling the take method because the data is reshuffled.)

[@jamesmccannon7769]

Are you sure that this is correct? When testing this although this is stopping the accuracy becoming 100%, it's validation accuracey is 1.00 from the start??

[@j220493]

There is another bug even bigger than that, he is doing data leakage. When he download the images from google, there are a lot of them duplicated or triplicated, so when he splits the data, same image is in all three divisions (train, val and test), for sure, so is its requeried to purge the data before all the process.

[@KarrsonHeumann]

I really love these longer tutorials. You explained things so well in this one that I feel like AI development finally clicked for me, not just in terms of this specific application, but also in general. I would understand if you'd be worried about length vs entertainment, but honestly you teach so well and you are so enthusiastic I don't think that should even be a concern. Thank you so much! :)

[@lemon_and_berry]

Agreed !! Waiting for such vids

[@pedrobizzotto556]

Its rare to see someone explain in detail every step of the way! Great tutorial!

[@andybrice2711]

But not _too_ much detail. It's a good balance of theory and practice.

[@anoushkachatterjee2922]

​@@andybrice2711 i am pretty much new to ml, i have an assignment on classifying images of equipments into defective and non defective. will i be able to do that if i follow this video?

[@andybrice2711]

​@@anoushkachatterjee2922 I don't think you need to follow this process if you just want to classify your own images. This video gets into the low-level details of building your own model from scratch. Probably what you would want to do is start with a pre-trained model (perhaps wrapped in a nice developer-friendly API) and then "fine tune" it. There might be something in OpenCV or YOLO. My advice would be to ask ChatGPT for recommendations.

[@atharvmunot8305]

Thank you so much for this Tutorial!! IT IS THE BEST !! P.S. A side note for the recent viewer, while compiling the model, use the command: model.compile('adam', loss = tf.losses.sparse_categorical_crossentropy, metrics = ['accuracy']) This change caters to the recent change in the naming conventions and ensures that the saved .h5 model runs when loaded

[@donguyengiac5046]

OMG you're a lifesaver!

[@Sneha-e7d]

@atharvmunot8305 hey!! Can you help me with installation.. actually I am having problem while installing the tensorflow gpu , its saying python version is not compatible , do you know any other way of installing that without degrading the python version.

[@hugehammer]

Wow! It was awesome. I built my first CNN architecture with the help of this video.

[@rhudinaormasa1962]

What environment did you use, vscode jupyter or Google colab or just jupyter?

[@salvinprasad8592]

Absolutely brilliant. I will use this structural approach in my third paper for my PhD. Thanks so much

[@adowanshahriar3623]

This tutorial is live savior. Recently I am doing my thesis on medical image processing and this video is an absolute guideline. Thanks a ton Nicholas :3

[@samarth2915]

yess !!! do u have any idea what changes should be done in the NN foro multi classes ??

[@priyam66]

@@samarth2915 for multiclass classification, following changes need to be made. 1) the activation function for the output layer in ANN will be Softmax 2) The loss function would be Categorical CrossEntropy(). 3) if you use this shown method of the data pipeline, then you will have to create multiple subfolders for each class in the multi-class classification problem.

[@edushineicscschool]

Can I get your number or mail id I'm also doing similar research in medical I need some clarification

[@TheSakyoGamer]

This. Was. AMAZING! Oh my gosh. Thank you for such for this tutorial. I've been wanting to get into machine learning for so long, but never knew where to start or how to work these models. With how long this video was and how excellent your commentary was, it helped so much! I plan to watch a ton of your videos about creating some more models.

[@dimasalangxt3482]

Amazing job on these videos! Would love to see a tutorial featuring 9 or more classes, thanks!

[@leonp5281]

I second that!!

[@qualitycontrol716]

please Nicholas

[@danielleivy8180]

??? I just kept adding classes, when it hits 9 it just moves onto 10....

[@poisonedexistence3027]

Yes, Nicholas please! multiclass tutorial from you is needed=) Thank you

[@mohamedgaal5340]

Thanks a lot Nick! I like how you skim through the mathematical concepts behind your code. Very informative! I'm watching the whole playlist :)

[@alextotheroh8071]

This is truly a fantastic tutorial. I had a working model in just a few hours. I didn't realize it could be done that quickly! Thank you!

[@prasadcode58]

Whatever I learned in theory, now learned in practical just because of you, it's really fun and put lots of efforts to make us understand in easier terms, thanks a lot. ✨

[@vigneshm4916]

Great Tutorial! As you said a tutorial on callbacks would be great. Thanks Nicholas!

[@NicholasRenotte]

Yeah, wish I spent some more time on it in this vid. You got it @Vignesh!

[@m1hz665]

Great tutorial bro , i had some theorical knowledge about CNN , and now for the fitst time you made me create for the first time a CNN model step by step , thank you so much , appreciate it bro , keep it up ❤💥

[@maximeentsi2205]

Best CNN tutorial I've never seen

[@prabanjan.a1750]

What CNN type is used here ?

[@binalweerasena753]

Now you don't need to pip install tensorflow-gpu. GPU features are already inside tensorflow main library. Trying to install tensorflow-gpu got me an error which ate up my time. Thanks Renotte for this amazing videos. You truly inspire us, the newbies.goodvibes

[@afkhoso]

thanks! had the same issue! comments and conversations like this help others very much!

[@TadiwanasheNyamapfene]

Thanks for commenting. I ran into the same error

[@no-name168]

39:46 awfully great moment. you watch some tutorial on youtube, you expect everything to be fine and then something like this happens, like it's not scripted anymore, watching him solving acutal problem. it's like 4th wall or something, love it

[@Christian-dd2qm]

Great content and I love that you speak proper English! I am not a native speaker and had my fill of Australian and Indian accents.

[@arnabghosh5547]

Nich, would please also make theory explaining CNN, object detection, their metrics & hyperparameter tuing

[@NicholasRenotte]

Ohhhh man, theory isn't really my fav but I'll see what I can do!

[@m.rishab4770]

​@@NicholasRenotte can You please make video on how to do it for multiple classes ?

[@dadadavid1409]

Yeah, please

[@russstephanson6987]

Nicholas this video is one of the best tutorials I have seen on image classification. Thank you

[@Alice8000]

Wow, I didn’t know Neon did programming videos too. You’re really smart. Clap 👏

[@mahendrakergaurav5867]

Amazing Tutorial, highly underrated channel, will share this with my friends.

[@eddyateye810]

I've learned more in 30mins than in my image processing class

[@yosephawoke9584]

As a student who is working on an image classification project, I learn a lot here and it was a very nice and interactive explantation. Thank You Nick!

[@somtoogbe108]

You are really a great teacher and I love the way you organize your code. Keep it up Nic

[@interlightstudio1645]

NIck has installed TF like 5,000 times by now loool. Gr8 tuts!!!

[@NicholasRenotte]

Legit, at least 5000 LOL

[@m.rishab4770]

​@@NicholasRenotte ​can You please make video on how to do it for multiple classes ?

[@ubaidabbas8175]

This was an amazing tut for a beginner like me. Thank you man... Great Explaination and Great Visualisation. Each part of your code was explained perfectly.

Hello Nick, thank you for this awesome tutorial, I learned a lot. I was wondering if you published another tutorial with more classes involved? (at 13:01) Thanks

[@bl4z3_kanazaki]

Dude makes DL actually fun to learn! I can't learn anything from the lecturers at my college because they talked to much and didn't even explain anything! Thanks man

[@Gordonias]

Would love to see some more stuff on deep reinforcement learning! :)

[@yusufkemaldemir9393]

Who are you bro:))) This is best training video I have seen on KZbin.

[@sourabhchandra1740]

Woow 😲 😁😄... Just amazing 😍.. what a video and superb explanation... Your explanation is top notch 😉... Got full idea on how to approach a machine learning project and confidence too!! Can you please make a separate video on "how to decide the architecture of a deep neural network" ?? That would be very helpful

[@talhasadaqat3742]

can you please let me know how you setup the environment , like how to install jupyter , and get the new file called image classification so that all the commands run perfectly without any errors

[@twopolaar599]

I just finished my model on classifying images of cats and dogs, and Im shocked at how accurate it is! The feeling of finally finishing it SO good, I finally feel accomplished.

[@malhargirgaonkar1668]

Can it work to classify image in more than 1 category? I have 11 categories

[@twopolaar599]

@@malhargirgaonkar1668 Yeah, but im pretty sure you got to adjust some of the code. For examples, you cant do anything with binary classification if you have more than two categories. I recommend trying to do it with two categories first, then add more categories later once you have finished

[@sanjanatarekar5942]

Wow!! This is the best tutorial. Thank you for making this. Please do 1 with multi classes classification, regularization, dropouts, normalization(basically tuning parameters) and confusion matric.😃

[@bratutub3]

Your detailed explanation has led me to a better understanding of the matter... Thank you...

[@felixzepol]

Amazing explanation, Im using this for thesis project, I'll let you know how well it went 👍

[@Nice_lolat]

Thanks man, exactly how i will like to learn. Everypart of the code explained and visualised. No assumption ☺

[@charanb1781]

Amazing tutorial !! Just diving into deep learning and your example came in handy.

[@fassoyangce1075]

Another amazing video! 👏👏👏

[@senaldharmasiri8990]

ay bro this is the best explanation i've found so far. Thanks

[@abdullahalmazed5387]

You have no idea how much you are helping me, Nick. Thaaaaaaanks!!! And those wondering if this tutorial is worth the time, I, as a beginner, can tell that this is a top-class tutorial for beginners in deep learning and CNN. Just go for it. Thanks again, Nick. PV: I learned the theories from an Andrew Ng course, and this tutorial helped me learn to implement those theories.

[@economicsanity2895]

yeah I agree. Initially I watched Nick's vids without any foundational knowledge, and I was utterly clueless as to what was going on. However, after acquiring some fundementals, it is absolutely great to understand it

[@peepdawg8995]

this is the most concise and precise video I have seen on KZbin regarding artificial intelligence. thank you very much sir, you're an excellent teacher.

[@mihai3678]

😂😂maybe because you are a beginner and have no idea what is a 'precise video ' about AI, ML

[@navarclaudio]

Massive video Nicholas!!! I'm very grateful!!

[@crisevangelene7242]

This video is a life-saver, thank you so much for sharing this video sir. Because of your help, me and my frnd could finish our project on time. A Big thankkkk youuuuuu from our side ❤️❤️❤️

[@dheerajkumarsiripalli9363]

And that friend is none other than me 😹 and once again thanks alot sir. Actually we have learned an outdated course which nearly ended our project... But your video saved us just on time. Thanks alotttt ❤️❤️❤️

[@RashmiBajaj-mf3el]

Hi, I implemented the code just as you. But at the end, it is not predicting happy and sad correctly. You only checked for one image. Even your code wouldn't work ig, I do not know what is the error, but you should correct it as well! Please le me know if you resolve it and how.

[@SoulFrmTitanic]

No, it is completely correct and it worked for me!

[@Aditya-ms1ll]

more epochs would work for you , or try changing optimizer and find suitable one asper you

[@ProjectSibanda]

Incredible Tutorial Nick!!

[@dariusdevo9554]

The best tf explanation I've ever seen, big thumb up!

[@MuhammadMuzammilAzad-q7y]

Never seen such a comprehensive tutorial.. just a beginner in ML and DL so such tutorials help alot.. thank you

[@lubwamaemmanuel7743]

Hey Nicholas, that is an amzing tutorial, i really learnt tonnes to take me to my next learning of ML. Thanks so much.💯

[@joelmaiza]

Realmente increíble, muy explicativo paso a paso y es de los pocos tutoriales que puedes seguir sin tener ninguna complicación. Gracias por compartir con todos.

[@FandangoJepZ]

would really appreciate one with more classes! Trying to make an AI for SET

[@aminaleali7161]

Exceptional talent for teaching! Informative, clear, and I love the pace of it. No fluff and to the point. Thank you and great job!

[@gaurav_0369]

it's good seeing a professional programmer debug a code and keep getting stuck...... feels relatable

[@leonp5281]

I would love to use more than 2 classes as you mentioned on 12:55. What do we need to change on the code ? Any possibility for an updated .ipynb file if you don't want to create another tutorial? Thank you so much for the great guides!!

[@pranavtiwari_yt]

have you figured out? plz tell if yes

[@marekkamm1524]

pls tell me too :) @@pranavtiwari_yt

[@RoccoEduxD2]

we need to see how to work with ore than 2 classes.

[@patricklynch6871]

You're the man Nicholas! Thanks for the video!

[@eru3890]

I love your videos, keep it up! I would like for you to make a video explaining about how to handle false positives with objects we don't want to detect.

[@jesseifrah5383]

holy shit this is absolutely incredible. the most amazing toutorial I have ever watched. thank you so much, and if there if any way I can pay you back, let me know sir! I don't even follow machine learning content, but I still subscribed to help out! thanks nicholas!

[@eduardozaratemontalvo6405]

MEN!! YOU ARE A HERO!! 🥳THANKS A LOT!! Very clean, Good Explanation. THANKS!!

[@Huapua]

A nice practical start to this topic. It makes me look forward to learning more of the details in order to troubleshoot and train correctly. Even though I was following along with Nicholas, my neural network was making incorrect predictions. I reran all my code from scratch and the same failed predictions. The third time I trained from scratch, it seems like the predictions were more likely to be accurate. It might be because my image downloader downloaded less images than Nicholas. I only had 3 batches of training data. I guess the point of all this is that if you are failing to get accurate predictions, maybe try rerunning your code to get different fit parameters, and/or get more data.

[@coding_with_chip]

just a heads up, typo: 1:09:50 - PART 4: Evaluating Perofmrnace

[@sanjayravi8181]

Hi Nick, You are right you were dividing the data twice by 255 so it came out to be 0.0039. (1/255 = 0.0039). Thanks for the video. Happy learning!.

[@kokylolify]

Finally some good tutorial, thank you Sir!

[@allisonmoczynski5655]

Thank you so much for making this tutorial! It was so, so helpful!

[@aysegulsezen4280]

It was very useful video. Thank you very much! This video answered my questions about preparing image input data for machine learning.

[@hassanbabaousmail194]

Woow, What a perfect explanation. Thank you so much for this tutorial.

[@josepalacios843]

The thumbnail says "This is proably terrible!" but the content hits like boulder. Nice work man. Thanks for sharing your experience.

[@brunospfc8511]

Thanks Nicholas, i'll try it

[@rmarti5163]

@1:06:00 I guess I still don't understand how the training is working. How does it know if it is correct or not? what is the data loss? is it the data that doesn't actually show a happy or sad human face?

[@archeacnos]

I've only watched the first 10 minutes and it's already so useful

[@lancemarchetti8673]

Nice! Imagine if we could build a classifier that can spot Base64 in a screen capture and extract it accordingly. In digital Forensics this could be quite handy in cases where base encoding is used to hide particular image data.

[@FahadRashid1100]

That weird colouring at 1:17:19 is because cv2 is expecting `uint8` and we're giving it `int`... so yeah... this works too but just saying... it was bugging me lol... Other than that, AMAZING tutorial! Life saver!

[@GamingShiiep]

58:00. I wish you had shown how not only the functions work, but how the data changes. Maybe you/someone here can explain to me: when specifying the number or kernels in conv2D, does that mean: one kernel does it job and creates a new matrix/np.array, then another random kernel does its job and creates a second new matrix/no array etc. And after that's happened 16 times, all these matrizes are getting stacked up or summed and meaned and then one single resulting matrix is fed into the next step which is max pooling? In max pooling basically the one matrix fed into it goes through convolution again and then the resolution gets lowered again by doing the max pooling kernel thing? Once again, summing up all 32 matrizes resulting in one meaned matrix? Or: the conv 2D puts out 16 new matrizes. Then for each matrix, max pooling is applied still putting out 16 matrizes, which are basically lowered in resolution. After that, those 16 matrizes are treated like the input was (single image matrix), which was to create n matrizes (in conv2D) out of a single input. In this case now it's creating 32 matrizes out of Number 1 matrix from the previous max pooling step. You see, i just don't understand the dimensions and in between outputs. That's why I really dislike piping in tutorials.

[@faizansiddiqui4901]

Nick, thank you so much for the valuable tutorial. really appreciated. 👍

[@photorealm]

Awesome video. Love the way you explained all of the steps in great common sense detail. 5 Stars 😊

[@ziddimalikziddimalik1288]

Dear sir, Your video is so awesome and you deliver each point very clearly and it need more video related this topics and student want to be more learn to your channel I hope you will be share more video such kind of work... Good job sir👍

[@khenpahilanga9596]

Nicholas, THAAAANK youuu!!

[@dragway0385]

Amazing, you are my hero Nich, bless you ❤

[@numforfrank9205]

Your videos are top notch, explicit and yet humorous at the same time😅. YOu make learning AI easy. Thanks Nic.

[@pritamrajbhar9504]

it was an amazing explanation, glad I visited this channel.

[@efeaydemir1413]

Brilliant!!!!!! Man thanks a lot, not finished yet. Allthough it is awesome so farr, learned a lot.

[@chetankumar5634]

@Nicholas Renotte Can you make a video more more than 2 classifier?

[@ameer-alahmadi]

The great explanation I've ever seen! Thanks a lot!

[@martinology5215]

Sir, this...was...amazing. Thank You! ✨

[@_leonard_6173]

45:50 - I am confused. What do you mean by "shuffle" ? I have the data just like yours : Data folder containing a folder for each class. Also another question that I have is : if my Data folder contains class A (50% of parent folder), class B (50% of parent folder), and I make the training set 70% of the data folder just like your code does, will it be 70% of each child folder or 70% of the entire Data folder?

[@eatonasher3398]

@46:00 Where is the "shuffle" function called? Wouldn't it shuffle each time you take a partition? If so, what good does skipping the first partitions do?

[@bibinlouis4927]

This video made me happy. Thank you sensei.

[@TheFranklinns]

thanks for the tutorial....well explained, i tried it and it's working perfect. Thanks Nicholas

[@PenAndSpecs007]

Amazing tutorial! Thanks Nicholas

[@cyclone760]

40:50 I felt it man. Shows he's just a human developer like us.

[@christophercrager1925]

What an amazing video. Extremely helpful. Thank you!

[@NicholasRenotte]

Thanks a mil @Christopher!

[@christophercrager1925]

@@NicholasRenotte have you ever thought of doing a video on beginner topics like K-means, PCA, TSNE and applying them to a real world example? I’d love to hear you talk about stuff like that

[@ridvanakyuz_]

Thank you ! This video so instructive and explanatory . But i need a help. Where we used to transfer learning ? Can you explain me please ?

[@prasadk6510]

This is excellent to classify any two classes of objects.But why we get error in multi-class however.Mismatch in input and output class.

[@CHARLYFIVE7]

It's almost the same. You just need to change some parameters and the activation function in the output.

[@lemon_and_berry]

Amazing video !! , enjoyed every second of it

[@ahmedbahpullo8544]

C'est ce que je voulais savoir Thank you you are the best, 💪💪💪

[@NicholasRenotte]

Thanks a mil @Ahmed!

[@abhishekfnu7455]

Thank you so much for this video. You really explained every bit of it.

[@danielleivy8180]

I need a good tutorial on ML....oh, Nick has a video No wait but I want to use YOLO....hey, Nick has a video... Now I need to figure out how to make my own custom LARGE dataset, everyone keeps pointing to established collections and I need to make my own....wait.....what....NICK HAS A VIDEO!!! 💃🏾💃🏾💃🏾👏🏾👏🏾👏🏾

[@ankitapal9229]

very useful and explanation is also just awesome..Thanks for the tutorial..