def __init__(self, ilist=None, shape=SHAPE):
self.transform = transforms.Compose([transforms.ToTensor()])
self.ilist = ilist
self.shape = shape
self.parsedata = ParseData()
return None
def __init__(self, ifile=IMAGE_LIST, sfile=SETTINGS):
self.data = pd.read_csv(ifile)
self.len = len(self.data)
self.parsedata = ParseData()
self.badindex = []
with open(sfile) as fp:
content = json.load(fp)
self.shape = content['shape']
self.servo_pred = NNTools(content["servo_setting"])
self.motor_pred = NNTools(content["motor_setting"])
self.servo_pred.load_model(content['servo_model'])
self.motor_pred.load_model(content['motor_model'])
return None
def beginAccTest(self):
data = pd.read_csv(self.TEST_LIST)
parsedata = ParseData()
with open(self.SETTINGS) as fp:
content = json.load(fp)
shape = content['shape']
servo_pred = NNTools(content["servo_setting"])
servo_pred.load_model(content['servo_model'])
servo_count = 0
for index in range(len(data)):
_, servo, motor = parsedata.parse_data(data["image"][index])
pred_servo = servo_pred.predict(data["image"][index])
if abs(servo - pred_servo) <= content['error_margin']:
# print(servo)
servo_count += 1
totalAcc = (100 * servo_count / (index + 1))
return totalAcc
class Datagen(Dataset):
#--------------------------------------------------------------------------
# method: constructor
#
# arguments:
# dfile:
#
# return: none
#
def __init__(self, ilist=None, shape=SHAPE):
self.transform = transforms.Compose([transforms.ToTensor()])
self.ilist = ilist
self.shape = shape
self.parsedata = ParseData()
return None
#
# end of method
#--------------------------------------------------------------------------
# method: get_image
#
# arguments:
# ifile: input image file
#
# return:
# image: image tensor
#
# This method returns image tensor
#
def get_image(self, ifile):
image = self.parsedata.parse_image(ifile)
image = cv2.resize(image, (self.shape[0], self.shape[1]), \
interpolation=cv2.INTER_CUBIC)
image = self.transform(image)
return image.unsqueeze(0)
#
# end of method
#--------------------------------------------------------------------------
# method: getitem
#
# arguments:
# index: index for list of images
#
# return: image, servo, motor
#
# This method returns data
#
def __getitem__(self, index):
# parse image, servo and motor data
image, servo, motor = self.parsedata.parse_data(self.ilist[index])
# convert image, servo and motor data to tensor
image = cv2.resize(image, (self.shape[0], self.shape[1]), \
interpolation=cv2.INTER_CUBIC)
image = self.transform(image)
servo = torch.tensor(float(servo))
motor = torch.tensor(float(motor))
return image, servo, motor
#
# end of method
#--------------------------------------------------------------------------
# method: length
#
# arguments: none
#
# return: none
#
# This method returns length of data
#
def __len__(self):
return len(self.ilist)
#
# end of method
#------------------------------------------------------------------------------
# Debugging Block ANI717
#------------------------------------------------------------------------------
#from torch.utils.data import DataLoader
#
#file = 'data/list/list_1.csv'
#
#dataloader = DataLoader(dataset=Datagen(file,shape=[10,8]), batch_size=1000, \
# shuffle=True)
#for image, servo, motor in dataloader:
# print(image.shape)
# print(servo, motor)
import json # used for settings parsing
import pandas as pd # used to parse csv
import time # used to caclulate prediction speed
# local files for neural net
from av_nn_tools import NNTools
from av_parse_data import ParseData
# where we store the csv relative to script
TEST_LIST = 'data/list/final_test.csv'
# where we store settings json relative to script
SETTINGS = 'data/set_accuracy_test.json'
# read the csv and parse it for prediction
data = pd.read_csv(TEST_LIST)
parsedata = ParseData()
# load settings from json file
with open(SETTINGS) as fp:
content = json.load(fp)
shape = content['shape']
servo_pred = NNTools(content["servo_setting"])
servo_pred.load_model(content['servo_model'])
servo_count = 0
accum_time = 0
# make predictions from csv and print value
for index in range(len(data)):
_, servo, motor = parsedata.parse_data(data["image"][index])
start_time = time.time()
pred_servo = servo_pred.predict(data["image"][index])
class DataTest:
#--------------------------------------------------------------------------
# method: constructor
#
# arguments:
# dfile: image data holder file
# dwidth: width of the images
# dheight: height of the images
#
# return: none
#
def __init__(self, ifile=IMAGE_LIST, sfile=SETTINGS):
self.data = pd.read_csv(ifile)
self.len = len(self.data)
self.parsedata = ParseData()
self.badindex = []
with open(sfile) as fp:
content = json.load(fp)
self.shape = content['shape']
self.servo_pred = NNTools(content["servo_setting"])
self.motor_pred = NNTools(content["motor_setting"])
self.servo_pred.load_model(content['servo_model'])
self.motor_pred.load_model(content['motor_model'])
return None
#
# end of method
#--------------------------------------------------------------------------
# method: display_pred
#
# arguments:
# index: index value
#
# return: none
#
# This method compares prediction with given data
#
def display_pred(self, index=0):
# parse image, servo and motor data
image, servo, motor = self.parsedata.parse_data(
self.data["image"][index])
# make prediction
pred_servo = self.servo_pred.predict(self.data["image"][index])
pred_motor = self.motor_pred.predict(self.data["image"][index])
# process image with servo and motor values for displey
image = cv2.resize(image, (self.shape[0], self.shape[1]), \
interpolation=cv2.INTER_CUBIC)
cv2.line(image, (240, 280), (240 - 20*(15 - servo), 100), \
(255, 0, 0), 3)
cv2.line(image, (220, 280), (220 - 20*(15 - pred_servo), 100), \
(255, 255, 0), 3)
image = cv2.putText(image, str(pred_motor) + ":" + str(motor), \
(180, 310), cv2.FONT_HERSHEY_SIMPLEX, 1, \
(255,255,255), 2, cv2.LINE_AA)
# show image
cv2.imshow('picture {}'.format(index), cv2.cvtColor(image, \
cv2.COLOR_RGB2BGR))
cv2.moveWindow('picture {}'.format(index), 500, 500)
return None
#
# end of method
#--------------------------------------------------------------------------
# method: display
#
# arguments:
# index: index value
# key: provided key press
# nfile: new csv file to store refined images
#
# return: none
#
# This method displays images, also lets user to refine image set by
# deleting unwanted images
#
def display(self, nfile=NEW_LIST, index=0, key=SPACE_KEY):
# when 'd' is pressed, (index-1) value is appended in badindex array
# when 's' is pressed, all image data except deleted indices is saved
# to new CSV image data file
if (key == DEL_KEY):
self.badindex.append(index - 1)
cv2.destroyWindow('picture {}'.format(index - 1))
elif (key == SAVE_KEY):
ref_data = self.data[0:index].drop(self.badindex)
ref_data.to_csv(nfile, index=False)
else:
cv2.destroyWindow('picture {}'.format(index - 1))
# parse image, servo and motor data
image, servo, motor = self.parsedata.parse_data(
self.data["image"][index])
# process image with servo and motor values for displey
image = cv2.resize(image, (self.shape[0], self.shape[1]), \
interpolation=cv2.INTER_CUBIC)
cv2.line(image, (240, 300), (240 - 20*(15 - servo), 200), \
(255, 0, 0), 3)
image = cv2.putText(image, str(motor), (180, 310), \
cv2.FONT_HERSHEY_SIMPLEX, \
1, (255,255,255), 2, cv2.LINE_AA)
# show image
cv2.imshow('picture {}'.format(index), cv2.cvtColor(image, \
cv2.COLOR_RGB2BGR))
cv2.moveWindow('picture {}'.format(index), 500, 500)
return None
#
# end of method
#------------------------------------------------------------------------------
# Debugging Block ANI717
#------------------------------------------------------------------------------
#import timeit
#QUIT_KEY = ord('q')
#total_start = timeit.default_timer()
#
#cardata = CarDataFile()
#for i in range(100):
# cardata.display_pred(index=i)
#
#total_stop = timeit.default_timer()
#print('Total time required: %f' % (total_stop - total_start))