def predict(img):
#predict the class of the input image
execute_net(features.get_HoG(img)) #Execute net base don the input image
pos = np.argmax(receptors[depth]) #Find the class having max. weight
print receptors[depth], dataset.folders[
pos] #Print the folder name corresponding to the class
return dataset.folder[pos]
def read_from_folder(path, val):
imgs = []
for f in os.listdir(path):
ext = os.path.splitext(f)[1]
if ext.lower() not in valid_images:
continue
filename = path+'/'+f
img = cv2.imread(filename,0)
feat = features.get_HoG(img)
imgs.append([feat, val])
return imgs
def read_from_folder(path, val):
#Read all the images in the folder
imgs = []
for f in os.listdir(path): #list all the files in the folder
ext = os.path.splitext(f)[1] #get the file extension
if ext.lower() not in valid_images: #check if the extension is valid for the image
continue
filename = path+'/'+f #create the path of the image
img = cv2.imread(filename,0) #read the image
if img == None: #Display error if image has nothing
print 'Error! Blank Image : ' + filename
continue
feat = features.get_HoG(img) #Get the features for the image
imgs.append([feat, val]) #append features with the categorical output
return imgs
def read_from_folder(path, val):
#Read all the images in the folder
imgs = []
for f in os.listdir(path): #list all the files in the folder
ext = os.path.splitext(f)[1] #get the file extension
if ext.lower(
) not in valid_images: #check if the extension is valid for the image
continue
filename = path + '/' + f #create the path of the image
img = cv2.imread(filename, 0) #read the image
if img == None: #Display error if image has nothing
print 'Error! Blank Image : ' + filename
continue
feat = features.get_HoG(img) #Get the features for the image
imgs.append([feat, val]) #append features with the categorical output
return imgs
def convert2requiredFormat(data_c):
data = []
imgs = data_c.get('data')
for i in range(len(imgs)):
val = data_c.get('fine_labels')[i]
img = data_c.get('data')[i]
img = img.reshape((3,32,32))
dst = np.zeros((32,32))
for x in range(32):
for y in range(32):
dst[x][y] = (img[0][x][y]+img[1][x][y]+img[2][x][y])/3
feat = features.get_HoG(dst)
if val == 85:
data.append([feat, [0,1]])
else:
data.append([feat, [1,0]])
return data
def convert2requiredFormat(data_c):
data = []
imgs = data_c.get('data')
for i in range(len(imgs)):
val = data_c.get('fine_labels')[i]
img = data_c.get('data')[i]
img = img.reshape((3, 32, 32))
dst = np.zeros((32, 32))
for x in range(32):
for y in range(32):
dst[x][y] = (img[0][x][y] + img[1][x][y] + img[2][x][y]) / 3
feat = features.get_HoG(dst)
if val == 85:
data.append([feat, [0, 1]])
else:
data.append([feat, [1, 0]])
return data
def execute(inputs):
global receptors
if inputs.size > 10:
start = time.clock()
receptors[0] = features.get_HoG(inputs) #compute the feature vector of the input image
print 'HoG time : ', (time.clock()-start)
start = time.clock()
for index in xrange(0,depth): #Execute Neural Network
receptors[index+1] = act.activate(synapses[index].dot(receptors[index]) + bias[index+1], False, act_fn[index+1])
print 'NNet time : ', (time.clock()-start)
pos = np.argmax(receptors[depth]) #Get the position of the maximum value output
if pos == 1:
if category[pos] > 0.9: #return 'Tracked' only if more than 90% sure
return category[pos]
return category[0]
def execute(inputs):
global receptors
if inputs.size > 10:
start = time.clock()
receptors[0] = features.get_HoG(
inputs) #compute the feature vector of the input image
print 'HoG time : ', (time.clock() - start)
start = time.clock()
for index in xrange(0, depth): #Execute Neural Network
receptors[index + 1] = act.activate(
synapses[index].dot(receptors[index]) + bias[index + 1], False,
act_fn[index + 1])
print 'NNet time : ', (time.clock() - start)
pos = np.argmax(
receptors[depth]) #Get the position of the maximum value output
if pos == 1:
if category[
pos] > 0.9: #return 'Tracked' only if more than 90% sure
return category[pos]
return category[0]
def predict(img):
#predict the class of the input image
execute_net(features.get_HoG(img)) #Execute net base don the input image
pos = np.argmax(receptors[depth]) #Find the class having max. weight
print receptors[depth], dataset.folders[pos] #Print the folder name corresponding to the class
return dataset.folder[pos]
def predict(img):
execute_net(features.get_HoG(img))
print receptors[depth]
pos = np.argmax(receptors[depth])
print dataset.folders[pos]