def create_validation_file(dataset_path):
# dataset_path = "./filtered-filtered-dataset-twoLane-Normalized/"
validation_percent = 0.1
validation_directory = "Validation"
#reset the directory - If you wanna use this first move the pictures back because it doesn't do that automatically
if not gfile.Exists(dataset_path + validation_directory):
# gfile.DeleteRecursively(dataset_path + validation_directory)
gfile.MakeDirs(dataset_path + validation_directory)
#get all the images inside the directory
files = os.listdir(dataset_path)
image_files = filterImages(files)
image_files = addDatasetPath(dataset_path, image_files)
#split it into validation files
validation_amount = int(validation_percent * len(image_files))
#this will randomly pop an x amount of labels
validation_files = [image_files.pop(random.randrange(len(image_files))) for _ in range(validation_amount)]
#move the validation files to a new folder
for name in validation_files:
#move images
name_split = name.split("/")
new_dest = dataset_path + validation_directory + "/" + name_split[-1]
os.rename(name, new_dest)
print("moved: ", name, "to", new_dest)
#move text files
extension_split = os.path.splitext(name_split[-1])
text_file_src_name = dataset_path + extension_split[0] + ".txt"
new_txt_dest = dataset_path + validation_directory + "/" + extension_split[0] + ".txt"
os.rename(text_file_src_name, new_txt_dest)
print("moved: ", text_file_src_name, "to", new_txt_dest)
print("---------------------")
# def filterImages(array): # newArr = [] # for name in array: # listName = os.path.splitext(name) # extension = listName[-1] # if extension == ".jpg": # newArr.append(name) # return newArr ###Begin main function here### # create an input tensor prepare_file_system(summaries_dir) #load the data set # A vector of filenames. image_files = os.listdir(dataset_path) image_files = filterImages(image_files) random.shuffle(image_files) #make the dataset the full path image_files = addDatasetPath(dataset_path, image_files) #files variable will be left with only the train ground truth train_groundTruth = extractLabels(image_files) #load validation fileset validation_files = os.listdir(validation_path) validation_files = filterImages(validation_files) validation_files = addDatasetPath(validation_path, validation_files) random.shuffle(validation_files) validation_groundTruth = extractLabels(validation_files) assert len(train_groundTruth) == len(image_files)
'''Reads through all the images in a folder one by one to find a corrupted image'''
import os
import tensorflow as tf
from dataHelperFunctions import filterImages
dataset_path = "./Full dataset - normalized/"
all_images = os.listdir(dataset_path)
all_images = filterImages(all_images)
name = tf.placeholder(tf.string)
image_string = tf.read_file(name)
image_decoded = tf.image.decode_jpeg(image_string)
with tf.Session() as sess:
for imageName in all_images:
print(imageName)
read_image = sess.run(image_decoded,
feed_dict={name: dataset_path + imageName})
logits, endpoints = mobileNet_v3.mobilenet(images, output_regression_count)
# Restore using exponential moving average since it produces (1.5-2%) higher
# accuracy
# ema = tf.train.ExponentialMovingAverage(0.999)o
# vars = ema.variables_to_restore()
# load the normalization params if the data is normalized
if normalized_dataset:
text_file = open(normalized_file_name, "r")
normalization_params = text_file.read().split()
text_file.close()
saver = tf.train.Saver()
file_names = os.listdir(dataset_path + validation_path)
file_names = filterImages(file_names)
file_names = addDatasetPath(dataset_path + validation_path, file_names)
# shuffle(file_names)
labels = extractLabels(file_names)
with tf.Session() as sess:
saver.restore(sess, checkpoint)
for i in range(len(file_names)):
x = logits.eval(feed_dict={file_input: file_names[i]})
print(x)
x = x[0]
unnormalizeFromParams(x, normalization_params)
image = cv2.imread(file_names[i])
image = drawOneLane(image, x[0], x[1], x[2], "blue")
import os
import statistics
import shutil
from dataHelperFunctions import filterImages, addDatasetPath, extractLabels
def normalize(data, mean, stdev):
return str((data - mean) / stdev)
dataset_path = "D:/cuLane/culane_preprocessing/converted_dataset_percentage_augmented/" #don't forget the slash at the end!
new_dataset_path = "D:/LaneDetectionV2/d_aug_two_lanes_percentage_dataset/"
print("reading data...")
files = os.listdir(dataset_path)
image_files = filterImages(files)
image_files = addDatasetPath(dataset_path, image_files) #merging the full path with the image file names in order to copy them properly later
labels = extractLabels(image_files)
print("read all data...")
#calculate the mean of each
meansOfAllData = []
stdevOfAllData = []
for i in range(len(labels[0])): #assuming that all the length of data will be constant
print("calculating mean and stddev at index: ", i)
allDataAcrossSingleIndex = []
for j in range(len(labels)):
allDataAcrossSingleIndex.append(float(labels[j][i]))
mean = statistics.mean(allDataAcrossSingleIndex)
std = statistics.stdev(allDataAcrossSingleIndex)