Feeding your own data set into the CNN model in Keras 

Fantastic video (and your others as well). Wish I had found this months ago.

Hi Anuj , It is a very helpful video for Deep Learning practitioners ... Thanks for the good work.. You enlightened me...Thanks , thanks a lot...

Almost everything i was hoping to find was right here in this videos. Thanks a ton!

Very nice video tutorial! Many thanks! Ran with a different data set of my own, very good results! :)

Beautiful! Sir, your work is very commendable! Thank you so much :)

You are my saviour Sir ! looking out for the exact explanation . Thanks a lot Sir and keep up your great service . God Bless

great tutorial Anuj. Thank you!

Thank you for your video Mr Anuj~

I need to ask if it is possible to feed pairs of image to the Keras and merge the result at the end or not

Hi Anuj, I have the dataset which has both cats and dogs in the same image. I have labeled it using bounding box techniques, I have XML file of the dimensions and converted it to csv. How can i push it for training? I want to use object detection API for the same. Please help

Thank you so much for this tutorial.

Please help......You have flatten the image matrix first, but may be convolution requires image still to be in matrix format than after that flattening of image is done (in model) perhaps. So how is this going to affect the model ?

I have to categorize dog breeding based on images of dogs. It will be done with the same way like the video?

sir, if iam trying to load the pretrainned model of alexnet i found some error , plz hlp me to fix it. error: ValueError: Layer #0 (named "conv2d_1" in the current model) was found to correspond to layer conv_1 in the save file. However the new layer conv2d_1 expects 1 weights, but the saved weights have 2 elements.

Sir could you please tell me how to give names to that images if I have very bulky

Man! You saved my career! I love you!

have images been downloaded in jpeg format ?

Can you tell me after convert the data into "categorical" how can i use it for real time prediction.... I am a student so can you please help me ?

Nice Video (y) In my case, I have database with images with random labels, Is there any way to arrange the labels or can we train with images having random labels?

may i know what version your models use? i mean your libraries, numpy and so on

i tried your code working good given right predict value when i used image from data/horse/horse.1.jpg , but when i tried to predict with one downloaded Horse image from Downloads it could n't be predict it ? why ? what was the issue . why it couldn't predict different location of image ? Please help me

HI anuj sir, I want to ask you one question .. How I can read 3D DICOM images?

Thank you for writing this. I just wanted to use my own images without downloading a data set. Could you make one make a csv dataset

hi Anuj, can you tell me how we can feed our own dataset(1D signal) to CNN??

Good work Anuj !! Thanks for helping out !! Wish you Great Success !!

hi anuj! do you have a project on facial keypoints detection using tensor flow ...???i will pay u..plz help me

Hi ANuj I am getting this error : AssertionError: AbstractConv2d Theano optimization failed: there is no implementation available supporting the requested options. Did you exclude both "conv_dnn" and "conv_gemm" from the optimizer? If on GPU, is cuDNN available and does the GPU support it? If on CPU, do you have a BLAS library installed Theano can link against? I have given my own single dataset of 80 finger print images and calculating losses.Can you please advise.

great work anuj..keep on...

At 19:30 I can see 2 Dropout() functions being called. First it is 0.25 and then below it is 0.5. So since you run the code altogether, the Dropout was at 0.5 at first itself. And the training accuracy didn't improve. And the second time when the second model.fit is executed, the split causes the improvement??

how to extract the features of testing dataset ???? i have tried like this but no working get_5th_layer_testoutput = K.function([model.layers[0].input, K.learning_phase()], [model.layers[10].output]) layer_testoutput5 = get_5th_layer_testoutput([test, 0])[0] plz hlp me how to extract testing data feature. i have extracted th training data features in the same manner & got good result.but iam failed in case of testing dataset

the immatrix in my computer is showing some error. Traceback (most recent call last): File "", line 3, in for im2 in imlist], dtype= float) ValueError: setting an array element with a sequence Can u help me out?

How can i test a text captcha... i mean how will i define classes for that? Captcha is combination of alphanumeric values, how can i test ir through CNN. Please help.

very good explaning : but i get this error when implement cnn models ((ValueError: ('The specified size contains a dimension with value

Hi Very use full tutorial you have here. However, What is the reason for this error when I run your code i got following error after run few epochs "have shapes {} and {}".format(x.shape, y.shape)) ValueError: x and y must have same first dimension, but have shapes (20,) and (3,)

hi.. i wanna ask.. you don't use tensorflow right in this part 1 tutorial?

you saved my day i was struggling with folders of 10 different classification images........

Hello Anuj, When I compile the "model = Sequential()" section I got an error: ValueError: ('The specified size contains a dimension with value

Sir can you please tell what modifications are to be done if I want to run code with three different folders separately

Is there a reason why you made a column of ones?

I would like to ask when you create the label array and you gave them numbers for each class if someone wants to give the actual names of each class what has to do? after that need to be one hot encoder? Thank you very much

how to extract the feature of individual layer

Hi Anuj shah, I have this error: "C:\Users\dr_h_\Anaconda2\envs\kersa_env\lib\site-packages\ipykernel\iostream.py", line 317, in write self._buffer.write(string) ValueError: I/O operation on closed file sometimes it appear some times not. I do not know how to handle this error and continue executing my program and read csv files Thanks

Hey, I got this Error, Because of the New Update of Keras ow to Fix This? UserWarning: The "show_accuracy" argument is deprecated, instead you should pass the "accuracy" metric to the model at compile time: `model.compile(optimizer, loss, metrics=["accuracy"])` warnings.warn('The "show_accuracy" argument is deprecated, ' train_acc=hist.history['accu'] KeyError: 'accu'

thanks Anuj...one question..will it be same for any other classification as well...I am new to it..so if u could brief me

thanks man this was really helpful!

Hi! Its a very nice video for beginners like me! Very helpful! However,this is about image classification. I am doing a project on lane detection and In my problem set, I have got a number of " road" images. Now,I want them to convert into gray scale and detect edges in it.The ultimate task is to detect the lane markings in all those images and later annotate them to a video. I was able to do this for all 3900 images and run the video(with the lanes detected) using computer vision(open cv). Now,whatever code I had used in open cv to detect the lanes in a single image,can the same code be used here by CNN? As in,I want to load all my images(resized) into CNN and then perform the image pre-processing and edge detection as per my open cv code and then get the output with the lanes detected in every image. Is it possible to do so in keras? i.e,using the same code as in computer vision approach in CNN? Please let me know.

Great resource!

how to prepare dataset for 1D signals instead of images?

I got follwing key error. Kindly help me with it TypeError Traceback (most recent call last) in () 4 model.add(Conv2D(nb_filters, (nb_conv, nb_conv)), 5 padding ='valid', ----> 6 input_shape=(1, None, None)) 7 convout1 = Activation('relu') 8 model.add(convout1) TypeError: add() got an unexpected keyword argument 'padding'

how can i get ur code for sample,good

im getting error as filter must not be larger than input filter(3,3) input (1,200), can u help me with this

Hi, Anuj.. What if the dataset has 3 channels color rgb? So, PIL image convert to what? Then, how is the flatten matrix in RGB? And the input shape in RGB? Thanks, your help will be so important :)

thanks for posting this video! It is just what I am looking for. I am running this code on Mac. However, I got the an error for the code below: for file in listing: im = Image.open(path1 + '\\' + file) img = im.resize((img_rows,img_cols)) gray = img.convert('L') #need to do some more processing here gray.save(path2 +'\\' + file, "JPEG") The error message shows: OSError: [Errno 2] No such file or directory: '\\Users\\huinaxu\\Downloads\\photos\\daisy' What could be the reason? Why it is double back slash here? Shall I modify the code for Mac, so that the path is single slash?

how can I use the same code to feed my own RGB images(3 channels) to keras?

Hey, how can we save our training model? Since i didnt see it on this tutorial

Very Nice presentation ******** dear how can it be possible that i don't load all of images to array means i don't have much memory and i have 10k images of 300x300 and i want to do it for RGB image

Hi Anuj, I get a file permissions error while prescribing and opening the input and output folders. Which IDE did you use to run the code, am trying from Anaconda but getting few errors

very good tutorial.

Hi Anuj, Thank you for video. Did you have experience of usage image preprocessing before feeding them in CNN? in case yes, which features did you extracted? could you please share a relevant link on some examples? Thank you.

Anuj do you have/plan any video or code on image segmentation to detected multiple objects in a single image? I found one in tensorflow (github.com/bhabanism/models/tree/master/object_detection) but looking for one in Keras.

Why do you expand the dimension at x = np.expand_dims(x, axis=0) ?

Hello again Anuj shah, I would like to ask for an Image classification problem. I use Corel5K (I do not know if you know that dataset?) I have achieved a low percentage (app. 30%-35% accuracy) how can I increase that percentage. The dataset has 50 classes with 100 images for each class.

Anuj which is better to use in fit(), the validation_data or validation_split?

Hi Anuj shah , GREAT tutorial! Im using 2 classes with 150 pictures each. everything was working fine. I had an error with OverFlowError, but I changed the line input shape to have not (1,row,col) but to (row,col,1) that was ok after. I am getting this error now when running model.fit 20:20 in your video. Exception: Error when checking model input: expected convolution2d_input_3 to have shape (None, 200, 200, 1) but got array with shape (241L, 1L, 200L, 200L). I cant find solution anywhere it seems. Any thoughts? :(

Good one.Thanks

Anuj sir can you design a cnn architecture that will take two inputs separately using Functional API model..... Feeding one input at a time using Sequential model is everywhere.... So can you explain using Functional API?

Thanks Anuj, great video. Would u please ceate a video on data frame.corr() and VIF???

good presentation and understanding but how can i fix this erroor my keras version is 2.3.1 AttributeError: module 'keras.backend' has no attribute 'set_image_dim_ordering'

muy bueno, bien explicado!!!

Thank you so much sir...

Hi Anuj, Fantastic tutorial. One question, while running your updated_custom_data_cnn.py from your git repository, I'm getting this error. File "/home/guest/anaconda3/lib/python3.7/site-packages/keras/engine/training_utils.py", line 131, in standardize_input_data 'with shape ' + str(data_shape)) ValueError: Error when checking input: expected conv2d_1_input to have 4 dimensions, but got array with shape (646, 128, 128) Could you please help? Thanks a lot.

very nice video tutorial. could you please tell me how i can select appropriate optimizer for my system. thank you

Thank you!

Any python code for running that with webcam so that i can use this in real life applications?

Hi Anuj, I am trying to use Keras for sentence classification. I have performed word2vec using gensim. I am confused about how to proceed now. How to give the output of word2vec as input?

Hey Anuj, Thanks for the tutorial. However I followed your steps but didn't get a high accuracy on my 5 classes training. I guess maybe for the labeling we need to use one-hot encoding but I don't know how to. (Figuring it out). Do you have any ideas?

Is it possible to do semantic segmentation with the code from this video?

thanks you! :D

nicely explained, unfortunatly i'm getting error in visualization part AttributeError: 'Conv2D' object has no attribute 'get_output' can you help me with this even after using .output() TypeError: 'Tensor' object is not callable is coming can you help me with this thank you

hello anuj, how can we use only deep select features of train convolutional network. i want to remove last fully connected layer, whioch is output layer.

Hi Anuj, Instead of doing the gray scale, I am using the colored images and the image dimension is different. i am using image dimension 120x120 with colored images and the shape of the image is immatrix[0].shape =(43200,) which is 120x120x3. so when i am going to reshape it later, I am getting error which is "" ValueError: total size of new array must be unchanged"". How can i handle this error so i can reshape it??

At first i'd like to thank you for this awesome tutorial. but i have a question: --> how do could i use multiple input folders to read in my data? sorry i am absolute newbie in programming.

Great Work

how to execute it in colab

i want to know how to set img_row,Img_col, does it vary with the size of dataset?

how can I download the data set that you have been used in the tutorials

excellent.

Fantastic video Anuj !! I have a small question, can this be used to train on a set of individual images and then finding the no of occurrences of that individual image within a bigger image. In other words, object detection in a larger image. Thanks

Its a good explanation. I have two questions though. 1. In line 45 why do you have to flattened the data while we can simply call the flatten function provided by keras when we are feeding to fully connected layer ? 2. I would like to know why you had converted all the labels to 1 in line 48. ?

Thanks Anuj, would you please explain why we need to reshape data before feeding to CNN? How the data look before and after reshape. A small example will help a alot

I have a question like How to use CNN for Feature Extraction...suppose I have 3 set of folders (for example purpose)..each folder contain 100 images...I need to extract the out put of last hidden layer(for each images i trained using CNN)...Ultimately I should get all the features of 300 images...then I will feed this data to another classifier ....pls help me in this regard

Is Keras available for windows?

Why didnt you do feature extraction for this problem statement? When do we use feature extraction then?

You have explained pin to pin up to Model Start, i am impressed . Can you help me with video link where did you explain about model creation steps step way step like this video???. that helps me a lot.

Helo sir, Your work is excellent. Can you please suggest a solution for classifying multiclass unlabelled images into respective class folders using a pretrained model.

Hi Anuj! Many thanks for this tutorial! Is there any updated version, as of 06.2018? Thanks!

can u please provide me the code in the video

thank you very much ...this video is worth watching and this channel is worth subscribing....

Hi Anuj, thank you for your wonderful tutorials. I implement it but I got error in part 6 which that is related to [ val_acc and val_loss]. However, I got this error: train_acc=hist.history['acc'] KeyError: 'acc' I don't know what is the problem. Can you please help me??

getting an error while print test score and test accuracy "warnings.warn('The "show_accuracy" argument is deprecated, '" What to do now, ?