Image classification using CNN (CIFAR10 dataset) | Deep Learning Tutorial 24 (Tensorflow & Python)
ฝัง
- เผยแพร่เมื่อ 23 ต.ค. 2020
- In this video we will do small image classification using CIFAR10 dataset in tensorflow. We will use convolutional neural network for this image classification problem. First we will train a model using simple artificial neural network and then check how the performance looks like and then we will train a CNN and see how the model accuracy improves. This tutorial will help you understand why CNN is preferred over ANN for image classification.
Code: github.com/codebasics/deep-le...
Exercise: Scroll to the very end of above notebook. You will find exercise description and solution link
Do you want to learn technology from me? Check codebasics.io/?... for my affordable video courses.
Deep learning playlist: • Deep Learning With Ten...
Machine learning playlist : th-cam.com/users/playlist?list...
#cnn #cnnimageclassification #imageclassificationpython #cnnmodel #deeplearning #tensorflowimageclassification #pythonimageclassification
🌎 My Website For Video Courses: codebasics.io/?...
Need help building software or data analytics and AI solutions? My company www.atliq.com/ can help. Click on the Contact button on that website.
#️⃣ Social Media #️⃣
🔗 Discord: / discord
📸 Dhaval's Personal Instagram: / dhavalsays
📸 Codebasics Instagram: / codebasicshub
🔊 Facebook: / codebasicshub
📱 Twitter: / codebasicshub
📝 Linkedin (Personal): / dhavalsays
📝 Linkedin (Codebasics): / codebasics
🔗 Patreon: www.patreon.com/codebasics?fa...
DISCLAIMER: All opinions expressed in this video are of my own and not that of my employers'.
Do you want to learn technology from me? Check codebasics.io/ for my affordable video courses.
sir plese plese reply i am doing a project on pcb defect detection using cnn model please help me out i am not getting it please help me
From Brazil, you are the best ML teacher!!! Thank you.
Thanks Luciano for your kind words
Excellent tutorials much better than many highly paid course floating online..Thanks a lot sir ..your videos helped me lot ...
Your tutorials are truly outstanding, surpassing many paid online courses. I want to express my deep appreciation for the invaluable support they've offered. Your detailed explanations of each code line have been incredibly helpful, particularly when I'm teaching machine learning to my students. Your videos provide a level of comprehension and utility that distinguishes them from other machine learning resources. Your efforts are greatly appreciated... Cheers!!!!!!!!!!!!!!!!💥💫💢
You are so much better than my university tutors :-D Thanks a lot for your help!
as you teach all concepts even a primary student can understand it easily. Seriously big fan of your teaching style
From South Korea, Learning Much Faster, Accurate than Univ. Thanks
🤗🤗🙏
We Asians are for us ❤
I am a young Ai and machine learning engineer from a IIIT and your videos are like food for me if i don't eat then I can't live .Great explanation ...
finally I commented after watching tons of your videos daily . Salute to your spirit sir you will reach 10 M subs soon cause AI and ML is growing exponentially and your videos in this direction in serving as no. 1 you tube channel for simple explanations on Practical AI,ML coding and more people will join with you soon and soon...
Ha ha .. thanks for your kind words of appreciation my friend :)
the important CNN concept is explained in superb and simple to understand , Thanks a lot
Thank you sir! Teaching is also a skill and you nailed it!
Indeed you are an excellent tutor. Your efforts are greatly appreciated .I am fun of you. I AM an AI and machine learning outreach ,you pave me the way .Thanks a lot for you support
Excellent tutorials much better than my professor! You are the best! thank you so much! your videos helped me a lot....
Exciting Times!! May this series long continue😁
yes it will. My goal is to cove all the topics and make this your one stop place for deep learning
I started to learn ml after getting inspirations from your videos. Thank you !
Happy to hear that sabrina!
@@codebasics lol lo
Plpp
Pl
@@codebasics pl
Very good explanation with a clear easily understandable video. Thank you for your tutorial. Loved it.
Amazing tutorial, thanks a lot for sharing! Saludos desde Argentina! 🇦🇷
You are the best teacher of mine. I'm grateful to you always. Thanks a lot, sir.
Zeenat, thanks for you kind words
REALLY A GOOD VIDEO , i finally understood implementing CNN using CIFAR10
You are really inspirational and have so much to idolize. Thank you!
Glad it was helpful!
great job sir.....keep making videos love to watch and learn from your videos
someone give this man a life elixir, he must give this knowledge for all the future generations
how come you tube did not recommend me this way before. Your videos are just perfect for people who want to learn Deep Learning and want to overcome the fear of AI
I love your way of teaching
Excellent content! Thank you very much.
Excellent demo, saved my time.
Such a Good Content.
I am really exciting for upcoming videos.
Glad to hear that
Thank you so much! Your tutorials are very helpful
Glad you like them!
All the way superb!!!! All videos.
Thank you a lot! You helped me with my project!
Glad it was helpful!
Thank you sir, excellent explanation
really good explanations. thanks for your great help
You are superb in teaching. Please make video on how to deploy such trained models to production.
Thank you, It is a great tutorial😍 on CNN
Mr. Modi (Mr. Patel) is one side and rest Opposition (Data Science TH-camr) is on the other side.
I really envy you (ONIDA TV) and you command that envy with your highest excellence.
I am a retired Sr. Citizen and love data science (not because I understand it) but because of the amazing things that Amazon and Tesla and Google are doing..
Please keep going..and may God give you a very long life..
thank you soo much. from the knowledge i gained from this video, i decided to also increase the number of epochs in the first network(ann) from 5 to 10 and that led to a slight increase in the training accuracy(0.49 to 0.54). and for the cnn i intentionally decided first use the SDG optimizer and later the adam which also gave two different but better results than the ann. i also adjusted the epochs in each case. this has given me some more ideas to play around with, with regards to this model. once again thank you for bn such a great teacher
me too i searched for my issue accuracy was 10 % and no increase however i increased hidden layers epochs , but what help me is changing the softmax to sigmoid and the number of hidden units it was 4 on my project here i found it 3000 , it increase my accuracy too , but based on what he choosed 3000 and 1000 hidden units ?
@@ahmedhelal920 More hidden units will recognize more patterns and more features, which will help if your images have many patterns and objects. It is always recommended to use more hidden units on layers and decrease it after every layer to reach a better solution.
Excellent explanation. 👏
Hi Thank you for all your tremendous work you make fall in love with Machine learning. don't you dare to stop;) Thank you so so so much.
Thanks for your kind words khan ☺️ and yes now after reading your comment I am not going to stop 😉
@@codebasics bless you.
Thank you for the efforts you put in all these vedios, it is giving us a clear image of what is happening in each part. Thanks alot
Thank you so much for detailed tutorial. Can you please make a video on Object detection? Specially Faster RCNN and Yolo models.
Tq so munch sir for continuing this series amazing content supreb nice explantion
You're most welcome sathiya
very nicely explained brother. Loved the teaching style and followed the explanation
😊😊👍
Excellent video. Thank you
I am happy this was helpful to you.
Thank you so much for this great tutorial. It is really helpful. I have a question, you used 'sparse crossentropy' in prediction and it's supposed to return the class number but the output of y_pred is an array of the probability of each class, and to get the predicted class we used argmax function to get the index of maximum value?
Hello Bassem, "Sparse categorical cross entropy" is the loss function to be used when the actual output Y in the dataset is not in the one hot encoded format. And, sir has used "softmax" as the final activation function in the code showed in the video. It is because of this function that the final output, y_pred is an array of the probability of each class. Hence, finally in order to get the index position of the maximum probability value, which is typically the output class predicted by the CNN model, sir has used the np.argmax function.
Thanks a lot for your great courses, is it possible for you to explain my question? How should we add non-image features to our CNN model (features like cat and dog prices) to our flatten layer? Does the CNN model new added features belong to which input image?
Your videos are very good...you explain every line of code...it really helps me a lot to teach ML to my students...your videos are even more useful then other ML videos...👌😊
Glad you like them!
Thank you very much sir for this 😊
Tq u 💯 much sir, this video is very helpful.😍❤️🌹👍🥰🇮🇳
Awesome really like the face to face introduction
Glad you like it
great sir thank you
Very lucid explanation
very nice videos thank you so much bro :)
sir I am really appreciated, the way you teach all the concepts related to CNN, and how to build it,
sir how can get more accuracy using Keras tuner, please make a video on that.
thank you very much sir
Can you suggest some good final year project ideas related to image classification.
I'll be grateful
GREAT SIR
Brilliant!
You are the best by far
I am happy this was helpful to you.
Hello, can you please guide for the K-NN, MLP, CNN, Decision Tree, K-Mean Clustering, regression to solve this CIFAR-10 dataset problem. And compare the accuracies for each of the methodologies used.
Great Work! keep it up!
Thanks a lot!
how to split the image data into training and testing in folders
I’m from Taiwan. It’s really helpful
Glad it was helpful!
Thanks for nice explanation. Easy to understand the concepts. Can you make video for region CNN and faster R CNN?
I want to use the weights in the hardware upper model in the model. So how do I print out that weight (I'm a beginner)
Realy sir I like your teaching way
Thanks and welcome
You are doing an amazing work.. I really get intrest in ml after watching your video explanation..
Sir I'm work on project "image classification using deep neural network" The data set is *CIFAR 10*. Paper on which I'm working it already has 80.2% of accuracy . So by using deep neural network algorithms can I make accuracy beyond 80%
permission to learn sir. thanks you
awesome
Your classes are really beginner friendly and I have a doubt will adding more layers improves the accuracy
yes it might. you can try adding them. sometimes too many layers will overfit a model and while accuracy improves on training set, on test set it might perform poorly. You can use regularization techniques such as adding dropout layer to tackle these issues partially
Your approach is very well. You can explain the topics so well and easy to understand the complex topic.
Glad to hear that, I am happy this was helpful to you.
Nice tutorial sir. Can you create a chatbot using ANN? I would like to know how you will test that. Thanks!
thanks a lot sir for your explanation. i got accuracy of 98.97% using cnn model
Great job
Can you share with your code?
Hi, thanks for the clear explanation. I was wondering why you did not use softmax activation function in the last layer instead of sigmoid? As far as I know, softmax is preferred in multiclass problems (like in this case) and sigmoid is used for binary classification problems. Let me know and I appreciate your answer in advance.
Really nice video...its helped me lot...
I want you to start Audio, Video processing tutorial also because I like it your teaching skills.
Glad it was helpful!
I found something very important. When you reshape your y into 1 dimension, save it in a different variable and use the original one (2d) in the training and test process. Otherwise, the results change a lot
Why results change alot?
can i use a similar cnn for object recognition? I want to give multiple labels for each image and in the output i would need the bounding boxes and the corresponding predicted label. How to prepare the dataset accordingly if i were to implement a cnn implemented in the video?Or are there any other deep learning models i could build for this application?
No one in universe can teach like this
Thanks zain for your kind words
amazing
Very nice explanation on CNN....
how you can simplify such complex topics ? You must be having rich experience in this field...😊
sir how i convert the prediction into csv file with column names (filname and label)
convert to pandas dataframe and just execute - df.to_csv('df.csv')
Thank you 🙏
Glad it was helpful!
I have one doubt.. like here we are working for colored images , we have 3 channels RGB , so do we need filters also different for all the channels or there will be only 1 filter?
@codebasics why flattening again in model when reshape() is used to do it ??
Sir, one small doubt.. you said that we can use categorical_crossentropy when there is one hot encoded output pattern.. but in this example we used sparse_categorical_crossentropy , but still we used 10 output neurons and output was considered as max of that...which is like one hot encoding only right..
The 10 output neurons gives the probability of each true possibility and its value will be ranging in 0 - 1 and in order to get the index position of the maximum probability value, which is typically the output class predicted by the CNN model, we used the np.argmax function.
i got an answer from stackExchange -
If your Yi's are one-hot encoded, use categorical_crossentropy. Examples (for a 3-class classification): [1,0,0] , [0,1,0], [0,0,1]
But if your Yi's are integers, use sparse_categorical_crossentropy. Examples for above 3-class classification problem: [1] , [2], [3].
For complete explanation check this -
stats.stackexchange.com/questions/326065/cross-entropy-vs-sparse-cross-entropy-when-to-use-one-over-the-other
Great video. I was hoping you'd visualize the CNN kernels so we could see what they looked like. You specified 32 of them. Does this mean that all 32 are used in every image, and thus are meaningful in every case? that is, you won't have one what has a koala's eyes because the input images also include, say, rocks, buildings, and GPU cards?
Dear Sir, I have a data-frame with shape: 6500 rows and 146 cols. It is not a 3D data. How can I apply input_shape parameter to use CNN model?
Sir, i am observing this issue, which is on local gpu (NVIDIA rtx 3050) cnn models are giving very low accuracies, for example i implemented mnist in cpu(achieved 98-99% accuracy) and in gpu(achieved around 10-11% accuracy). for ann it works fine, i followed the same steps for integrating cuda and cudnn from the same website given by you. what to do?
Hi sir , based on what we can choose 3000 hidden units and 1000 hidden units on our project ?
Could you explain in detail about the reshaping process, on why its necessary ?
Great video thank you for your efforts in creating this , just a small doubt when I replicated the ANN model and ran the code without normalizing the data X_train and test Im getting 100% accuracy in train as well as test where as after normalizing it comes down to 50% and in this video you said then normalization is done to increase the accuracy then how is it happening? (Thank you in for your answer)
Thank you
model.evaluate:10000/10000 [==================] - 1s 57us/sample - loss: 0.0275 - accuracy: 0.9910
Sir, can you please show us how to plot the accuracy curve for the cnn model
Sir thanks for an amazing video. I am having a little trouble visualising the numpy array and how the pixel values are stored to eventually form the image, any video link you'd suggest for that please?
from keras.datasets import cifar10
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train[0] will give you the arrays as 0 is the row recall, 1 is the column.
Hi beautiful video! I have some special image in Black and white to be classified. I have two questions:
1. Do you think it is better to colorize them in order to improve the predicion?
2. If yes at the first question, what is a suitable technique to add colors?
Thanks a lot.
Not necessary as long as your test or prediction is also B&W
We should use softmax for multiclass classification right?. But here we used sigmoid? How is it executing?
i basically need to give an ai images and a numerical values, then i want to predict the numerical value from an image, is this kind of model suitable? what do you suggest?
how can you get more accuracy? I have messed around with the hyper parameters a lot but I can't seem to find something that gets me a good accuracy (above 80-85)
Thank you for the awesome tutorial. I have one question. Is there a way so I could give a path to one folder and then it would classify images which are in it using this model?
Yes you can use tensorflow dataset pipeline for that watch TF data pipeline tutorial in this same playlist
@@codebasics Thank You, I'll definitely watch it.
sir how to load the dataset if it is not available already in the Keras or TensorFlow library.
hi. im getting an error saying "AttributeError: module 'tensorflow' has no attribute 'Conv2D' " What do i do?
nice!!!
Thank you! Cheers!