3D CNN in Keras - Action Recognition

# The code for 3D CNN for Action Recognition
# Please refer to the youtube video for this lesson

3D CNN-Action Recognition Part-1

3D CNN-Action Recognition Part-2

from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation, Flatten
from keras.layers.convolutional import Convolution3D, MaxPooling3D

from keras.optimizers import SGD, RMSprop
from keras.utils import np_utils, generic_utils

import theano
import os
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import cv2
from sklearn.cross_validation import train_test_split
from sklearn import cross_validation
from sklearn import preprocessing


# image specification
img_rows,img_cols,img_depth=16,16,15


# Training data

X_tr=[]           # variable to store entire dataset

#Reading boxing action class

listing = os.listdir('kth dataset/boxing')

for vid in listing:
    vid = 'kth dataset/boxing/'+vid
    frames = []
    cap = cv2.VideoCapture(vid)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)
 

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
         frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()

    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)

#Reading hand clapping action class

listing2 = os.listdir('kth dataset/handclapping')

for vid2 in listing2:
    vid2 = 'kth dataset/handclapping/'+vid2
    frames = []
    cap = cv2.VideoCapture(vid2)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)

#Reading hand waving action class

listing3 = os.listdir('kth dataset/handwaving')

for vid3 in listing3:
    vid3 = 'kth dataset/handwaving/'+vid3
    frames = []
    cap = cv2.VideoCapture(vid3)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)

#Reading jogging action class

listing4 = os.listdir('kth dataset/jogging')

for vid4 in listing4:
    vid4 = 'kth dataset/jogging/'+vid4
    frames = []
    cap = cv2.VideoCapture(vid4)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)

#Reading running action class
 
listing5 = os.listdir('kth dataset/running')

for vid5 in listing5:
    vid5 = 'kth dataset/running/'+vid5
    frames = []
    cap = cv2.VideoCapture(vid5)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)
 

#Reading walking action class 

listing6 = os.listdir('kth dataset/walking')

for vid6 in listing6:
    vid6 = 'kth dataset/walking/'+vid6
    frames = []
    cap = cv2.VideoCapture(vid6)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)



X_tr_array = np.array(X_tr)   # convert the frames read into array

num_samples = len(X_tr_array)
print num_samples

#Assign Label to each class

label=np.ones((num_samples,),dtype = int)
label[0:100]= 0
label[100:199] = 1
label[199:299] = 2
label[299:399] = 3
label[399:499]= 4
label[499:] = 5


train_data = [X_tr_array,label]

(X_train, y_train) = (train_data[0],train_data[1])
print('X_Train shape:', X_train.shape)

train_set = np.zeros((num_samples, 1, img_rows,img_cols,img_depth))

for h in xrange(num_samples):
    train_set[h][0][:][:][:]=X_train[h,:,:,:]
 

patch_size = 15    # img_depth or number of frames used for each video

print(train_set.shape, 'train samples')

# CNN Training parameters

batch_size = 2
nb_classes = 6
nb_epoch =50

# convert class vectors to binary class matrices
Y_train = np_utils.to_categorical(y_train, nb_classes)


# number of convolutional filters to use at each layer
nb_filters = [32, 32]

# level of pooling to perform at each layer (POOL x POOL)
nb_pool = [3, 3]

# level of convolution to perform at each layer (CONV x CONV)
nb_conv = [5,5]

# Pre-processing

train_set = train_set.astype('float32')

train_set -= np.mean(train_set)

train_set /=np.max(train_set)



# Define model

model = Sequential()
model.add(Convolution3D(nb_filters[0],nb_depth=nb_conv[0], nb_row=nb_conv[0], nb_col=nb_conv[0], input_shape=(1, img_rows, img_cols, patch_size), activation='relu'))

model.add(MaxPooling3D(pool_size=(nb_pool[0], nb_pool[0], nb_pool[0])))

model.add(Dropout(0.5))

model.add(Flatten())

model.add(Dense(128, init='normal', activation='relu'))

model.add(Dropout(0.5))

model.add(Dense(nb_classes,init='normal'))

model.add(Activation('softmax'))

model.compile(loss='categorical_crossentropy', optimizer='RMSprop')

 
# Split the data

X_train_new, X_val_new, y_train_new,y_val_new =  train_test_split(train_set, Y_train, test_size=0.2, random_state=4)


# Train the model

hist = model.fit(X_train_new, y_train_new, validation_data=(X_val_new,y_val_new),
          batch_size=batch_size,nb_epoch = nb_epoch,show_accuracy=True,shuffle=True)


#hist = model.fit(train_set, Y_train, batch_size=batch_size,
#         nb_epoch=nb_epoch,validation_split=0.2, show_accuracy=True,
#           shuffle=True)


 # Evaluate the model
score = model.evaluate(X_val_new, y_val_new, batch_size=batch_size, show_accuracy=True)
print('Test score:', score[0])
print('Test accuracy:', score[1])



# Plot the results
train_loss=hist.history['loss']
val_loss=hist.history['val_loss']
train_acc=hist.history['acc']
val_acc=hist.history['val_acc']
xc=range(100)

plt.figure(1,figsize=(7,5))
plt.plot(xc,train_loss)
plt.plot(xc,val_loss)
plt.xlabel('num of Epochs')
plt.ylabel('loss')
plt.title('train_loss vs val_loss')
plt.grid(True)
plt.legend(['train','val'])
print plt.style.available # use bmh, classic,ggplot for big pictures
plt.style.use(['classic'])

plt.figure(2,figsize=(7,5))
plt.plot(xc,train_acc)
plt.plot(xc,val_acc)
plt.xlabel('num of Epochs')
plt.ylabel('accuracy')
plt.title('train_acc vs val_acc')
plt.grid(True)
plt.legend(['train','val'],loc=4)
#print plt.style.available # use bmh, classic,ggplot for big pictures
plt.style.use(['classic'])

Feeding your own data set into the CNN model in Keras

# The code for Feeding your own data set into the CNN model in Keras
# please refer to the you tube video for this lesson -

https://www.youtube.com/watch?v=2pQOXjpO_u0&index=18&list=PLd9i_xMMzZF7eIjnVuPxggYuGPg5CnA1l

####just copy and paste the below given code to your shell

#KERAS
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation, Flatten
from keras.layers.convolutional import Convolution2D, MaxPooling2D
from keras.optimizers import SGD,RMSprop,adam
from keras.utils import np_utils

import numpy as np
import matplotlib.pyplot as plt
import matplotlib
import os
import theano
from PIL import Image
from numpy import *
# SKLEARN
from sklearn.utils import shuffle
from sklearn.cross_validation import train_test_split

# input image dimensions
img_rows, img_cols = 200, 200

# number of channels
img_channels = 1

#%%
#  data

path1 = 'C:\Users\Ripul\Documents\Python Scripts\keeras-cnn-tutorial\input_data'    #path of folder of images    
path2 = 'C:\Users\Ripul\Documents\Python Scripts\keeras-cnn-tutorial\input_data_resized'  #path of folder to save images    

listing = os.listdir(path1)
num_samples=size(listing)
print num_samples

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")

imlist = os.listdir(path2)

im1 = array(Image.open('input_data_resized' + '\\'+ imlist[0])) # open one image to get size
m,n = im1.shape[0:2] # get the size of the images
imnbr = len(imlist) # get the number of images

# create matrix to store all flattened images
immatrix = array([array(Image.open('input_data_resized'+ '\\' + im2)).flatten()
              for im2 in imlist],'f')
               
label=np.ones((num_samples,),dtype = int)
label[0:89]=0
label[89:187]=1
label[187:]=2


data,Label = shuffle(immatrix,label, random_state=2)
train_data = [data,Label]

img=immatrix[167].reshape(img_rows,img_cols)
plt.imshow(img)
plt.imshow(img,cmap='gray')
print (train_data[0].shape)
print (train_data[1].shape)

#%%

#batch_size to train
batch_size = 32
# number of output classes
nb_classes = 3
# number of epochs to train
nb_epoch = 20


# number of convolutional filters to use
nb_filters = 32
# size of pooling area for max pooling
nb_pool = 2
# convolution kernel size
nb_conv = 3

#%%
(X, y) = (train_data[0],train_data[1])


# STEP 1: split X and y into training and testing sets

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=4)


X_train = X_train.reshape(X_train.shape[0], 1, img_rows, img_cols)
X_test = X_test.reshape(X_test.shape[0], 1, img_rows, img_cols)

X_train = X_train.astype('float32')
X_test = X_test.astype('float32')

X_train /= 255
X_test /= 255

print('X_train shape:', X_train.shape)
print(X_train.shape[0], 'train samples')
print(X_test.shape[0], 'test samples')

# convert class vectors to binary class matrices
Y_train = np_utils.to_categorical(y_train, nb_classes)
Y_test = np_utils.to_categorical(y_test, nb_classes)

i = 100
plt.imshow(X_train[i, 0], interpolation='nearest')
print("label : ", Y_train[i,:])

#%%

model = Sequential()

model.add(Convolution2D(nb_filters, nb_conv, nb_conv,
                        border_mode='valid',
                        input_shape=(1, img_rows, img_cols)))
convout1 = Activation('relu')
model.add(convout1)
model.add(Convolution2D(nb_filters, nb_conv, nb_conv))
convout2 = Activation('relu')
model.add(convout2)
model.add(MaxPooling2D(pool_size=(nb_pool, nb_pool)))
model.add(Dropout(0.5))

model.add(Flatten())
model.add(Dense(128))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(nb_classes))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adadelta')

#%%

hist = model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,
              show_accuracy=True, verbose=1, validation_data=(X_test, Y_test))
           
           
hist = model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,
              show_accuracy=True, verbose=1, validation_split=0.2)


# visualizing losses and accuracy

train_loss=hist.history['loss']
val_loss=hist.history['val_loss']
train_acc=hist.history['acc']
val_acc=hist.history['val_acc']
xc=range(nb_epoch)

plt.figure(1,figsize=(7,5))
plt.plot(xc,train_loss)
plt.plot(xc,val_loss)
plt.xlabel('num of Epochs')
plt.ylabel('loss')
plt.title('train_loss vs val_loss')
plt.grid(True)
plt.legend(['train','val'])
print plt.style.available # use bmh, classic,ggplot for big pictures
plt.style.use(['classic'])

plt.figure(2,figsize=(7,5))
plt.plot(xc,train_acc)
plt.plot(xc,val_acc)
plt.xlabel('num of Epochs')
plt.ylabel('accuracy')
plt.title('train_acc vs val_acc')
plt.grid(True)
plt.legend(['train','val'],loc=4)
#print plt.style.available # use bmh, classic,ggplot for big pictures
plt.style.use(['classic'])




#%%      

score = model.evaluate(X_test, Y_test, show_accuracy=True, verbose=0)
print('Test score:', score[0])
print('Test accuracy:', score[1])
print(model.predict_classes(X_test[1:5]))
print(Y_test[1:5])



#%%

# visualizing intermediate layers

output_layer = model.layers[1].get_output()
output_fn = theano.function([model.layers[0].get_input()], output_layer)

# the input image

input_image=X_train[0:1,:,:,:]
print(input_image.shape)

plt.imshow(input_image[0,0,:,:],cmap ='gray')
plt.imshow(input_image[0,0,:,:])


output_image = output_fn(input_image)
print(output_image.shape)

# Rearrange dimension so we can plot the result 
output_image = np.rollaxis(np.rollaxis(output_image, 3, 1), 3, 1)
print(output_image.shape)


fig=plt.figure(figsize=(8,8))
for i in range(32):
    ax = fig.add_subplot(6, 6, i+1)
    #ax.imshow(output_image[0,:,:,i],interpolation='nearest' ) #to see the first filter
    ax.imshow(output_image[0,:,:,i],cmap=matplotlib.cm.gray)
    plt.xticks(np.array([]))
    plt.yticks(np.array([]))
    plt.tight_layout()
plt

# Confusion Matrix

from sklearn.metrics import classification_report,confusion_matrix

Y_pred = model.predict(X_test)
print(Y_pred)
y_pred = np.argmax(Y_pred, axis=1)
print(y_pred)
 
                       (or)

y_pred = model.predict_classes(X_test)
print(y_pred)

p=model.predict_proba(X_test) # to predict probability

target_names = ['class 0(BIKES)', 'class 1(CARS)', 'class 2(HORSES)']
print(classification_report(np.argmax(Y_test,axis=1), y_pred,target_names=target_names))
print(confusion_matrix(np.argmax(Y_test,axis=1), y_pred))

# saving weights

fname = "weights-Test-CNN.hdf5"
model.save_weights(fname,overwrite=True)



# Loading weights

fname = "weights-Test-CNN.hdf5"
model.load_weights(fname)


# please refer to the you tube video for this lesson -

https://www.youtube.com/watch?v=2pQOXjpO_u0&index=18&list=PLd9i_xMMzZF7eIjnVuPxggYuGPg5CnA1l

tracking and detecting red colour object

% 1 is the default id of webcam
vid = videoinput('winvideo',1,'YUY2_640x480');

% Set the properties of the video object
set(vid, 'FramesPerTrigger', Inf);
set(vid, 'ReturnedColorspace', 'rgb')
vid.FrameGrabInterval = 5;

%start the video aquisition here
start(vid)

% Set a loop that stop after 400 frames of aquisition
while(vid.FramesAcquired<=400)
   
    % Get the snapshot of the current frame
    im = getsnapshot(vid);
   
    % Now to detect red objects in real time we have to subtract the red component layer
    % from the grayscale image to extract all the red objects in the image.
    im2 = imsubtract(data(:,:,1), rgb2gray(data));
    %Use a median filter to filter out noise in the image
    im3 = medfilt2(im2, [3 3]);
    % Convert the resulting grayscale image into a binary image.
    im4 = im2bw(im3,0.18);
   
    % Remove all those objects less than 300 pixels.

     im5 = bwareaopen(im4,300);
   
    % Label all the connected components in the image.
    bw = bwlabel(im5, 8);
   
  
    % We get a set of properties for each labeled region.
  stats=regionprops(bw,'BoundingBox','Centroid');
   
    % Display the image
    imshow(im)
   
    hold on
   
    %This is a loop to bound the red objects in a rectangular box.
    for object = 1:length(stats)
        bb = stats(object).BoundingBox;
        bc = stats(object).Centroid;
        rectangle('Position',bb,'EdgeColor','r','LineWidth',2)
        plot(bc(1),bc(2), '-m+')
        a=text(bc(1),bc(2), strcat('X: ', num2str(round(bc(1))), '    Y: ', num2str(round(bc(2)))));
        set(a, 'FontName', 'Arial', 'FontWeight', 'bold', 'FontSize', 12, 'Color', 'yellow');
    end
   
    hold off
end
% Both the loops end here.

% Stop the video aquisition.
stop(vid);

% Flush all the image data stored in the memory buffer.
flushdata(vid);

% Clear all variables
clear all