Tutorial on CNN implementation for own data set in keras(TF & Theano backend)-part-2 

can I have your contact number plz

@@UniqueTech0098 you can reach me in mail - anujonline645@gmail.com

Good luck

Thanks. I am glad my videos are useful

Its already mentioned - model.load_weights('weight_file_name')

Not exactly, this code is for single label

Keras has a preprocessing module for few standard architecture s but with opencv you have more options and flexibility.

yes sure

ues you can do. Mobilenet is a lightr weight model

@@anujshah645 from keras.preprocessing.image import ImageDataGenerator from keras.models import Sequential from keras.layers import Conv2D, MaxPooling2D from keras.layers import Activation ,Dropout,Flatten,Dense from keras import backend as K import tensorflow.compat.v1 as tf import numpy as np from keras.preprocessing import image from __future__ import generator_stop img_width,img_height=150,150 train_data_dir='data/train' validation_data_dir='data/validation' nb_train_samples=1000 nb_validation_samples=100 epochs=50 batch_size=20 if K.image_data_format()=='channels_first': input_shape=(3,img_width,img_height) else: input_shape=(img_width,img_height,3) train_datagen=ImageDataGenerator( rescale=1./255, shear_range=0.2, zoom_range=0.2, horizontal_flip=True) test_datagen=ImageDataGenerator(rescale=1./255) train_generator=train_datagen.flow_from_directory( train_data_dir, target_size=(img_width,img_height), batch_size=batch_size, class_mode='binary') validation_generator=test_datagen.flow_from_directory( validation_data_dir, target_size=(img_width,img_height), batch_size=batch_size, class_mode='binary' ) model=Sequential() model.add(Conv2D(32,(3,3),input_shape=input_shape)) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2,2))) model.summary() model.add(Conv2D(32,(3,3))) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2,2))) model.add(Conv2D(64,(3,3))) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2,2))) model.add(Flatten()) model.add(Dense(64)) model.add(Activation('relu')) model.add(Dropout(0.5)) model.add(Dense(1)) model.add(Activation('sigmoid')) model.summary() model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy']) model.fit_generator( train_generator, steps_per_epoch=nb_train_samples // batch_size, epochs=epochs, validation_data=validation_generator, validation_steps=nb_validation_samples // batch_size) i used above code but getting error RuntimeError: generator raised StopIteration , Can you help me ?

Hi Manjeet, The input to CNN is of shape - (num_samples,num_rows,num_cols,num_channels). If you are using gray image then image shape is usually say (128,128). It does not show you the channel. but yo u have to feed the channel info as well i.e (128,128,1). so your input_array shape should be (646,128,128,1) for gray images and (646,128,128,3) for coloured images. in you can do an expands dims function on 4th axis and a 1 will be added at the end. you can use this command x=np.expand_dims(x, axis=0)

Thanks Anuj for the reply. Waiting for more videos (GAN's , Ensemble methods etc.)

@@anujshah645 where i can plot that code x=np.expand_dims(x, axis=0)

@@ManjeetKaur-hm4jt where i can plot that code x=np.expand_dims(x, axis=0)

in which line is the error being generated, is it during testing?

I'm getting the same error, any solution to this?

+Poonam Zagade I guess in keras 2 we don't need the show accuracy argument

just use model.evaluate(X_test, Y_test, verbose=0)

well the error says for itself

Solved thanks

The model is not a perfect model, it was just a demo, you need to take better model, train it in a better way

@@anujshah645 why you say that I need your tetegram please

Can you post the issue

@@anujshah645 can I sent it via email?

@@anujshah645 the problem occurred when I executed this code: hist = model.fit(X_train, y_train, batch_size=16, epochs=num_epoch, verbose=1, validation_data=(X_test, y_test)) The error message is: "ValueError: Calling `Model.fit` in graph mode is not supported when the `Model` instance was constructed with eager mode enabled. Please construct your `Model` instance in graph mode or call `Model.fit` with eager mode enabled".

well you have to see the number of epoch that you selected for training and number of epoch that you gave during plotting

Thanks Anuj, It solved my bug..

that should be based on trial and error you can start with simple network first and see how it performs

thank you mister!

yes you can save the best model during training using callback function. i have explained it in this video

make your num_classes=40, the command will create one-hot encoding for it

thank you it works

well I am also not very sure about this error but you may check your protobuf installation, which tensorflow you are using and on which environment

Thanks Anuj. I don't know the exact problem as there is some problem within libraries [from their github discussion]. But, as soon as I restart the kernel of Spyder IDE, It worked fine.

Thats great. Good Luck!

There is some command to get the number of parameters for each layer in keras. I don't exactly remember that but you can get it online

OKay, Thanks for your reply.

+Wasif Khan similar like data folder you can make a test folder and load test data from there

thanks for the reply, but when I load test data. Should do all process the same you did for the training data? Because for testing data we dont need labels. Can you post some links to follow? thanks

Hi Prativa...Please add this command before splitting the input data (line 70 in the code) x=np.expand_dims(x, axis=1) I hope it will resolve the issue.

@@ManjeetKaur-hm4jt hi , i get ValueError: Input 0 is incompatible with layer conv2d_54: expected ndim=4, found ndim=5 after input x=np.expand_dims(x, axis=1) (line 70 in the code )

@@ManjeetKaur-hm4jt thankksss it's work for me

RnF Channel how did you solve it I am still getting the same error

change the backend from theano to tf in keras.json file. the code should work

Where is the keras.json file? Unable to get it

it should be in your home directory in .keras folder

Thankyou anuj I did that.Now, when I am running the testing line it says THEANO requires BLAS whereas I have changed my backend to tf and that prevents me from running the epoch.

check carefully or google to see how to change the backend

+Poonam Zagade what do you exactly mean by 3 D images. If you are referring to depth then we can use the depth as NUM o channel in the code above

I haven't tried but i guess you can

Check whether the images are RGB or gray already , you can check by computing their dim len(img.shape)

the prediction will always be in integer , as each image is mapped to a label. if you want to display name then simply use a if loop if predicted_label==0: name='cat' elif predicted_label==1: name = 'horse' and so on... yes my paypal is active: www.paypal.me/anujshah645

it is shown in the video , how to test any image after the model is trained using model.predict

Do you mean continuous output of predicted class from say webcam feeds. If so you can watch this video I have shown here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-G9qiGuZKjLs.html

I mean that the output instead of being classes will be continuous value, for example - I have images and I want to measure size of an object(of course I have training data)

oh I got it. you can treat as a regression problem , at the last layer instead of softmax you can just have linear activation layer

yes, I think this is the direction but I get dimensions error.. can we talk privately?

can I get ur email id...

i have explained it in the first video how to read the sub folders in the first video

anujonline645@gmail.com

Anuj shah thank you I will send u an email

the inpu should have shape - (num_samples, num_channels, num_rows, num_cols) for theano backend and (num_samples, num_rows, num_cols, num_channels) for tensorflow backend. In your case the num_channels dimension is missing , it should be 1 for gray image and 3 for colored image

@@anujshah645 thank you Sir