train.py

import tflearn
import cv2
from tflearn.data_preprocessing import ImagePreprocessing
from tflearn.data_augmentation import ImageAugmentation
import numpy as np

print "Loading data-----------------------"

k = 0

X = []
Y_t = []

for i in [1,2,3,4,5]:
	while True:
		img = cv2.imread('./'+str(i)+'/'+str(k)+'.jpg')
		if img == None:
			break
		X.append(cv2.resize(img.astype('float'),(100,100)))
		Y_t.append([i-1])
		k += 1
print X
print Y_t
Y = tflearn.data_utils.to_categorical(Y_t,5)
print Y
Y_t=[]

'''
image_prep = ImagePreprocessing()
image_prep.add_featurewise_zero_center()
image_prep.add_featurewise_stdnorm()

image_aug = ImageAugmentation()
image_aug.add_random_flip_leftright()
image_aug.add_random_rotation(max_angle=25.)
'''
print "Creating net----------------"
#t_norm=tflearn.initializations.uniform(minval=-1.0,maxval=1.0)

net = tflearn.input_data(shape = [None,100,100,3]
			#,data_preprocessing = image_prep)
			#,data_augmentation = image_aug
			)

net = tflearn.conv_2d(net,100,3 ,activation = 'ReLU')#,weights_init=t_norm)
net = tflearn.conv_2d(net, 80,3 ,activation = 'ReLU')#,weights_init=t_norm)
net = tflearn.max_pool_2d(net,2)

net = tflearn.conv_2d(net,75,3 ,activation = 'ReLU')#,weights_init=t_norm)
net = tflearn.conv_2d(net, 40,3 ,activation = 'ReLU')#,weights_init=t_norm)
net = tflearn.max_pool_2d(net,2)

net = tflearn.conv_2d(net,64,3 ,activation = 'ReLU')#,weights_init=t_norm)
net = tflearn.conv_2d(net,32,3 ,activation = 'ReLU')#,weights_init=t_norm)
net = tflearn.max_pool_2d(net,2)

net = tflearn.fully_connected(net, 200, activation = 'ReLU')#,weights_init=t_norm)
#net = tflearn.dropout(net,0.5)

net = tflearn.fully_connected(net, 5, activation = 'softmax')#,weights_init=t_norm)

rm = tflearn.optimizers.RMSprop (learning_rate=0.001, decay=0.5, momentum=0.2)

net = tflearn.regression(net, optimizer = rm,
			loss = 'categorical_crossentropy')

model = tflearn.DNN(net)

print "Net created------------------"

model.fit(X, Y, n_epoch=20, shuffle=True,
	  snapshot_epoch=False,
          batch_size=10)

print np.argmax(model.predict(X[:15]))+1
print Y[:15]

model.save('test.tflearn')

cam = cv2.VideoCapture(0)

while True:
	ret_val, img = cam.read()
	cv2.imshow('my webcam', img)
	img = cv2.resize(img.astype('float'), (100, 100)) 
	k = cv2.waitKey(1)
	if k==27:
		break
	elif k==32:
		print np.argmax(model.predict([img]))
		print model.predict([img])

cv2.destroyAllWindows()