I've built a neural network with keras using the mnist dataset and now I'm trying to use it on photos of actual handwritten digits. Of course I don't expect the results to be perfect but the results I currently get have a lot of room for improvement.
For starters I test it with some photos of individual digits written in my clearest handwriting. They are square and they have the same dimensions and color as the images in the mnist dataset. They are saved in a folder called individual_test like this for example: 7(2)_digit.jpg.
The network often is terribly sure of the wrong result which I'll give you an example for:
The results I get for this picture are the following:
result: 3 . probabilities: [1.9963557196245318e-10, 7.241294497362105e-07, 0.02658148668706417, 0.9726449251174927, 2.5416460047722467e-08, 2.6078915027483163e-08, 0.00019745019380934536, 4.8302300825753264e-08, 0.0005754049634560943, 2.8358477788259506e-09]
So the network is 97% sure this is a 3 and this picture is by far not the only case. Out of 38 pictures only 16 were correctly recognised. What shocks me is the fact that the network is so sure of its result although it couldn't be farther from the correct result.
EDIT
After adding a threshold to prepare_image (img = cv2.threshold(img, 0.1, 1, cv2.THRESH_BINARY_INV)[1]) the performance has slightly improved. It now gets 19 out of 38 pictures right but for some images including the one shown above it still is pretty sure of the wrong result. This is what I get now:
result: 3 . probabilities: [1.0909866760000497e-11, 1.1584616004256532e-06, 0.27739930152893066, 0.7221096158027649, 1.900260038212309e-08, 6.555900711191498e-08, 4.479645940591581e-05, 6.455550760620099e-07, 0.0004443934594746679, 1.0013242457418414e-09]
So it now is only 72% sure of its result which is better but still ...
What can I do to improve the performance? Can I prepare my images better? Or should I add my own images to the training data? And if so, how would I do such a thing?
EDIT
This is what the picture displayed above looks like after applying prepare_image to it:
After using threshold this is what the same picture looks like:
In comparison: This is one of the pictures provided by the mnist dataset:
They look fairly similar to me. How can I improve this?
Here's my code (including threshold):
# import keras and the MNIST dataset
from tensorflow.keras.datasets import mnist
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from keras.utils import np_utils
# numpy is necessary since keras uses numpy arrays
import numpy as np
# imports for pictures
import matplotlib.pyplot as plt
import PIL
import cv2
# imports for tests
import random
import os
class mnist_network():
def __init__(self):
""" load data, create and train model """
# load data
(X_train, y_train), (X_test, y_test) = mnist.load_data()
# flatten 28*28 images to a 784 vector for each image
num_pixels = X_train.shape[1] * X_train.shape[2]
X_train = X_train.reshape((X_train.shape[0], num_pixels)).astype('float32')
X_test = X_test.reshape((X_test.shape[0], num_pixels)).astype('float32')
# normalize inputs from 0-255 to 0-1
X_train = X_train / 255
X_test = X_test / 255
# one hot encode outputs
y_train = np_utils.to_categorical(y_train)
y_test = np_utils.to_categorical(y_test)
num_classes = y_test.shape[1]
# create model
self.model = Sequential()
self.model.add(Dense(num_pixels, input_dim=num_pixels, kernel_initializer='normal', activation='relu'))
self.model.add(Dense(num_classes, kernel_initializer='normal', activation='softmax'))
# Compile model
self.model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
# train the model
self.model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=10, batch_size=200, verbose=2)
self.train_img = X_train
self.train_res = y_train
self.test_img = X_test
self.test_res = y_test
def predict_result(self, img, show = False):
""" predicts the number in a picture (vector) """
assert type(img) == np.ndarray and img.shape == (784,)
if show:
img = img.reshape((28, 28))
# show the picture
plt.imshow(img, cmap='Greys')
plt.show()
img = img.reshape(img.shape[0] * img.shape[1])
num_pixels = img.shape[0]
# the actual number
res_number = np.argmax(self.model.predict(img.reshape(-1,num_pixels)), axis = 1)
# the probabilities
res_probabilities = self.model.predict(img.reshape(-1,num_pixels))
return (res_number[0], res_probabilities.tolist()[0]) # we only need the first element since they only have one
def prepare_image(self, img, show = False):
""" prepares the partial images used in partial_img_rec by transforming them
into numpy arrays that the network will be able to process """
# convert to greyscale
img = img.convert("L")
# rescale image to 28 *28 dimension
img = img.resize((28,28), PIL.Image.ANTIALIAS)
# inverse colors since the training images have a black background
#img = PIL.ImageOps.invert(img)
# transform to vector
img = np.asarray(img, "float32")
img = img / 255.
img[img < 0.5] = 0.
img = cv2.threshold(img, 0.1, 1, cv2.THRESH_BINARY_INV)[1]
if show:
plt.imshow(img, cmap = "Greys")
# flatten image to 28*28 = 784 vector
num_pixels = img.shape[0] * img.shape[1]
img = img.reshape(num_pixels)
return img
def partial_img_rec(self, image, upper_left, lower_right, results=[], show = False):
""" partial is a part of an image """
left_x, left_y = upper_left
right_x, right_y = lower_right
print("current test part: ", upper_left, lower_right)
print("results: ", results)
# condition to stop recursion: we've reached the full width of the picture
width, height = image.size
if right_x > width:
return results
partial = image.crop((left_x, left_y, right_x, right_y))
if show:
partial.show()
partial = self.prepare_image(partial)
step = height // 10
# is there a number in this part of the image?
res, prop = self.predict_result(partial)
print("result: ", res, ". probabilities: ", prop)
# only count this result if the network is at least 50% sure
if prop[res] >= 0.5:
results.append(res)
# step is 80% of the partial image's size (which is equivalent to the original image's height)
step = int(height * 0.8)
print("found valid result")
else:
# if there is no number found we take smaller steps
step = height // 20
print("step: ", step)
# recursive call with modified positions ( move on step variables )
return self.partial_img_rec(image, (left_x + step, left_y), (right_x + step, right_y), results = results)
def individual_digits(self, img):
""" uses partial_img_rec to predict individual digits in square images """
assert type(img) == PIL.JpegImagePlugin.JpegImageFile or type(img) == PIL.PngImagePlugin.PngImageFile or type(img) == PIL.Image.Image
return self.partial_img_rec(img, (0,0), (img.size[0], img.size[1]), results=[])
def test_individual_digits(self):
""" test partial_img_rec with some individual digits (shape: square)
saved in the folder 'individual_test' following the pattern 'number_digit.jpg' """
cnt_right, cnt_wrong = 0,0
folder_content = os.listdir(".\individual_test")
for imageName in folder_content:
# image file must be a jpg or png
assert imageName[-4:] == ".jpg" or imageName[-4:] == ".png"
correct_res = int(imageName[0])
image = PIL.Image.open(".\\individual_test\\" + imageName).convert("L")
# only square images in this test
if image.size[0] != image.size[1]:
print(imageName, " has the wrong proportions: ", image.size,". It has to be a square.")
continue
predicted_res = self.individual_digits(image)
if predicted_res == []:
print("No prediction possible for ", imageName)
else:
predicted_res = predicted_res[0]
if predicted_res != correct_res:
print("error in partial_img-rec! Predicted ", predicted_res, ". The correct result would have been ", correct_res)
cnt_wrong += 1
else:
cnt_right += 1
print("correctly predicted ",imageName)
print(cnt_right, " out of ", cnt_right + cnt_wrong," digits were correctly recognised. The success rate is therefore ", (cnt_right / (cnt_right + cnt_wrong)) * 100," %.")
def multiple_digits(self, img):
""" takes as input an image without unnecessary whitespace surrounding the digits """
#assert type(img) == myImage
width, height = img.size
# start with the first square part of the image
res_list = self.partial_img_rec(img, (0,0),(height ,height), results = [])
res_str = ""
for elem in res_list:
res_str += str(elem)
return res_str
def test_multiple_digits(self):
""" tests the function 'multiple_digits' using some images saved in the folder 'multi_test'.
These images contain multiple handwritten digits without much whitespac surrounding them.
The correct solutions are saved in the files' names followed by the characte '_'. """
cnt_right, cnt_wrong = 0,0
folder_content = os.listdir(".\multi_test")
for imageName in folder_content:
# image file must be a jpg or png
assert imageName[-4:] == ".jpg" or imageName[-4:] == ".png"
image = PIL.Image.open(".\\multi_test\\" + imageName).convert("L")
correct_res = imageName.split("_")[0]
predicted_res = self.multiple_digits(image)
if correct_res == predicted_res:
cnt_right += 1
else:
cnt_wrong += 1
print("Error in multiple_digits! The network predicted ", predicted_res, " but the correct result would have been ", correct_res)
print("The network predicted correctly ", cnt_right, " out of ", cnt_right + cnt_wrong, " pictures. That's a success rate of ", cnt_right / (cnt_right + cnt_wrong) * 100, "%.")
network = mnist_network()
# this is the image shown above
result = network.individual_digits(PIL.Image.open(".\individual_test\\7(2)_digit.jpg"))
Update:
You have three options to achive a better performance in this particular task:
Use Convolutional network as it performs better in tasks with spatial data, like images and are more generative classifier, like this one.
Use or Create and/or generate more pictures of your types and train your network with them your network to be able to learn them too.
Preprocess your images to be better aligned to the original MNIST images, against which you trained your network before.
I've just made an experiment. I checked the MNIST images regarding one represented number each. I took your images and made some preprocessing I proposed to you earlier like:
1. made some threshold, but just downwards eliminating the background noice because the original MNIST data has some minimal threshold only for the blank background:
image[image < 0.1] = 0.
2. Surprisingly the size of the number inside of the image has proved to be crucial, so I scaled the number inside of the 28 x 28 image e.g. we have more padding around the number.
3. I inverted the images as the MNIST data from keras has inverted also.
image = ImageOps.invert(image)
4. Finally scaled data with, as we did it at the training as well:
image = image / 255.
After the preprocessing I trained the model with MNIST dataset with the parameters epochs=12, batch_size=200 and the results:
Result: 1 with probabilities: 0.6844741106033325
result: **1** . probabilities: [2.0584749904628552e-07, 0.9875971674919128, 5.821426839247579e-06, 4.979299319529673e-07, 0.012240586802363396, 1.1566483948399764e-07, 2.382085284580171e-08, 0.00013023221981711686, 9.620113416985987e-08, 2.5273093342548236e-05]
Result: 6 with probabilities: 0.9221984148025513
result: 6 . probabilities: [9.130864782491699e-05, 1.8290626258021803e-07, 0.00020504613348748535, 2.1564576968557958e-07, 0.0002401985548203811, 0.04510130733251572, 0.9221984148025513, 1.9014490248991933e-07, 0.03216308355331421, 3.323434683011328e-08]
Result: 7 with probabilities: 0.7105212807655334
Note:
result: 7 . probabilities: [1.0372193770535887e-08, 7.988557626958936e-06, 0.00031014863634482026, 0.0056108818389475346, 2.434678014751057e-09, 3.2280522077599016e-07, 1.4190952857262573e-09, 0.9940618872642517, 1.612859932720312e-06, 7.102244126144797e-06]
Your number 9 was a bit tricky:
As I figured out the model with MNIST dataset picked up two main "features" regarding 9. Upper and lower parts. Upper parts with nice round shape, as on your image, is not a 9, but mostly 3 for your model trained against the MNIST dataset. Lower part of 9 is mostly a straighten curve as per the MNIST dataset. So basicly your perfect shaped 9 is always a 3 for your model because of the MNIST samples, unless you will train again the model with sufficiant amount of samples of your shaped 9. In order to check my thoughts I made a subexperiment with 9s:
My 9 with skewed upper parts (mostly OK for 9 as per MNIST) but with slightly curly bottom (Is not OK for 9 as per MNIST):
Result: 9 with probabilities: 0.5365301370620728
My 9 with skewed upper parts (mostly OK for 9 as per MNIST) and with straight bottom (Is OK for 9 as per MNIST):
Result: 9 with probabilities: 0.923724353313446
Your 9 with the misinterpreted shape properties:
Result: 3 with probabilities: 0.8158268928527832
result: 3 . probabilities: [9.367801249027252e-05, 3.9978775021154433e-05, 0.0001467708352720365, 0.8158268928527832, 0.0005801069783046842, 0.04391581565141678, 6.44062723154093e-08, 7.099170943547506e-06, 0.09051419794559479, 0.048875387758016586]
Finally just a proof for the image scaling (padding) importance what I mentioned as crucial above:
Result: 3 with probabilities: 0.9845736622810364
Result: 9 with probabilities: 0.923724353313446
So we can see that our model picked up some features, which it interprets, classifies always as 3 in the case of an oversized shape inside of the image with low padding size.
I think that we can get a better performance with CNN, but the way of sampling and preprocessing is always crucial for getting the best performance in an ML task.
I hope it helps.
Update 2:
I found another issue, what I checked as well and proved to be true, that the placement of number inside of image is crucial as well, which makes sense by this type of NN. A good example the number 7 and 9 which have been placed of center in MNIST dataset, near to bottom of the image resulted in harder or flase classification if we place the new number for classifying in the center of image. I checked the theory shifting the 7s and 9s towards to the bottom, so lefting more place at the top of the image and the result was almost 100% accuracy.
As this is a spatial type problem, I guess that, with CNN we could eliminate it with more effectiveness. However would be better, if MNIST was alligned to center, or we can do it programatically to avoid the issue.
What was your test score,on MNIST dataset?
And one thing that is coming to my mind that your images are missing thresholding,
Thresholding is a technique where the pixel value below a certain pixel is made to zero,See OpenCV thresholding examples anywhere,You probaly need to use inverse thresholding and check your results again.
Do,inform if there is some progress.
The main problem you have is that the images you are testing are different from the MNIST images, probably due to the preparation of images you have done, can you show an image from the ones you are testing with after that you apply prepare_image on it.
I want to use CNN network to segment 2 objects (binary: "0: object not present, 1: object present") into shapes but I have an issue with data. The train data is 150 images and in "jpg" format and the ground truth (label data) is also 150 images of "png" rasters of 0 and 1 (resulting in black white images).
Now the question is how to load this hybrid of train images and label images in Keras/Tensorflow and if there`s a dummy example and/or demonstration on how to do that in Python, I would be grateful.
You can define one generator for reading the input images and another one for reading the labels using the ImageDataGenerator class and its flow_from_directory() method, and then combine these two generators in a single generator. Just make sure the directory structure and (order of) file names of input and label images are the same:
data_image_gen = ImageDataGenerator(...)
data_label_gen = ImageDataGenerator(...)
image_gen = data_image_gen.flow_from_directory(image_directory,
# no need to return labels
class_mode=None,
# don't shuffle to have the same order as labels
shuffle=False)
image_gen = data_image_gen.flow_from_directory(label_directory,
color_mode='grayscale',
# no need to return labels
class_mode=None,
# don't shuffle to have the same order as images
shuffle=False)
def final_gen(image_gen, label_gen):
for data, labels in zip(image_gen, label_gen):
# divide labels by 255 to make them like masks i.e. 0 and 1
labels /= 255.
# remove the last axis, i.e. (batch_size, n_rows, n_cols, 1) --> (batch_size, n_rows, n_cols)
labels = np.squeeze(labels, axis=-1)
yield data, labels
# ... define your model
# fit the model
model.fit_generator(final_gen(image_gen, label_gen), ...)
Following this tutorial: https://www.tensorflow.org/versions/r1.3/get_started/mnist/pros
I wanted to solve a classification problem with labeled images by myself. Since I'm not using the MNIST database, I spent days creating my own dataset inside tensorflow. It looks like this:
#variables
batch_size = 50
dimension = 784
stages = 10
#step 1 read Dataset
filenames = tf.constant(filenamesList)
labels = tf.constant(labelsList)
#step 2 create Dataset
dataset = tf.data.Dataset.from_tensor_slices((filenames, labels))
#step 3: parse every image in the dataset using `map`
def _parse_function(filename, label):
#convert label to one-hot encoding
one_hot = tf.one_hot(label, stages)
#read image file
image_string = tf.read_file(filename)
image_decoded = tf.image.decode_image(image_string, channels=3)
image = tf.cast(image_decoded, tf.float32)
return image, one_hot
#step 4 final input tensor
dataset = dataset.map(_parse_function)
dataset = dataset.batch(batch_size) #batch_size = 100
iterator = dataset.make_one_shot_iterator()
images, labels = iterator.get_next()
images = tf.reshape(images, [batch_size,dimension]).eval()
labels = tf.reshape(labels, [batch_size,stages]).eval()
for _ in range(10):
dataset = dataset.shuffle(buffer_size = 100)
dataset = dataset.batch(batch_size)
iterator = dataset.make_one_shot_iterator()
images, labels = iterator.get_next()
images = tf.reshape(images, [batch_size,dimension]).eval()
labels = tf.reshape(labels, [batch_size,stages]).eval()
train_step.run(feed_dict={x: images, y_:labels})
Somehow using a higher batch_sizes will break python. What I'm trying to do is to train my neural network with new batches on each iteration. That's why Im also using dataset.shuffle(...). Using dataset.shuffle also breaks my Python.
What I wanted to do (because shuffle breaks) is to batch the whole dataset. By evaluating ('.eval()') I will get a numpy array. I will then shuffle the array with numpy.random.shuffle(images) and then pick up some the first elements to train it.
e.g.
for _ in range(1000):
images = tf.reshape(images, [batch_size,dimension]).eval()
labels = tf.reshape(labels, [batch_size,stages]).eval()
#shuffle
np.random.shuffle(images)
np.random.shuffle(labels)
train_step.run(feed_dict={x: images[0:train_size], y_:labels[0:train_size]})
But then here comes the problem that I can't batch the my whole dataset. It looks like that the data is too big for python to work with.
How should I solve this differently?
Since I'm not using the MNIST database there isn't a function like mnist.train.next_batch(100) which comes handy for me.
Notice how you call shuffle and batch inside your for loop? This is wrong. Datasets in TF work in the style of functional programming, so you are actually defining a pipeline for preprocessing the data to feed into your model. In a way, you give a recipe that answers the question "given this raw data, which operations (map, etc.) should I do to get batches that I can feed into my neural network?"
Now you are modifying that pipeline for every batch! What happens is that the first iteration, the batch size is, say [32 3600]. The next iteration, the elements of this shape are batched again, to [32 32 3600], and so on.
There's a great tutorial on the TF website where you can find out more how Datasets work, but here are a few suggestions how you can resolve your problem.
Move the shuffling to right after "Step 2" in your code. Then you are shuffling the whole dataset so your batches will have a good mixture of examples. Also increase the buffer_size argument, this works in a different way than you probably assume. It's usually a good idea to shuffle as early as possible, as it can be a slow operation if you have a large dataset -- the shuffled part of dataset will have to be read into memory. Here it does not really matter whether you shuffle the filenames and labels, or the read images and labels -- but the latter will have more work to do since the dataset is larger by that time.
Move batching and the iterator generator to be the last steps, just before starting your training loop.
Don't use feed_dict with Dataset iterators to input data into your model. Instead, define your model in terms of the outputs of iterator.get_next() and omit the feed_dict argument. See more details from this Q&A: Tensorflow: create minibatch from numpy array > 2 GB
Ive been getting through a lot of problems with creating tensorflow datasets. So I decided to use OpenCV to import images.
import opencv as cv
imgDataset = []
for i in range(len(files)):
imgDataset.append(cv2.imread(files[i]))
imgDataset = np.asarray(imgDataset)
the shape of imgDataset is (num_img, height, width, col_channels). Getting the i-th image should be imgDataset[i].
shuffling the dataset and getting only batches of it can be done like this:
from sklearn.utils import shuffle
X,y = shuffle(X, y)
X_feed = X[batch_size]
y_feed = y[batch_size]
Then you feed X_feed and y_feed into your model
Based on the Tensorflow tutorial for ConvNet, some points are not readily apparent to me:
are the images being distorted actually added to the pool of original images?
or are the distorted images used instead of the originals?
how many distorted images are being produced? (i.e. what augmentation factor was defined?)
The flow of functions for the tutorial seems to be as follows:
cifar_10_train.py
def train
"""Train CIFAR-10 for a number of steps."""
with tf.Graph().as_default():
[...]
# Get images and labels for CIFAR-10.
images, labels = cifar10.distorted_inputs()
[...]
cifar10.py
def distorted_inputs():
"""Construct distorted input for CIFAR training using the Reader ops.
Returns:
images: Images. 4D tensor of [batch_size, IMAGE_SIZE, IMAGE_SIZE, 3] size.
labels: Labels. 1D tensor of [batch_size] size.
Raises:
ValueError: If no data_dir
"""
if not FLAGS.data_dir:
raise ValueError('Please supply a data_dir')
data_dir = os.path.join(FLAGS.data_dir, 'cifar-10-batches-bin')
return cifar10_input.distorted_inputs(data_dir=data_dir,
batch_size=FLAGS.batch_size)
and finally cifar10_input.py
def distorted_inputs(data_dir, batch_size):
"""Construct distorted input for CIFAR training using the Reader ops.
Args:
data_dir: Path to the CIFAR-10 data directory.
batch_size: Number of images per batch.
Returns:
images: Images. 4D tensor of [batch_size, IMAGE_SIZE, IMAGE_SIZE, 3] size.
labels: Labels. 1D tensor of [batch_size] size.
"""
filenames = [os.path.join(data_dir, 'data_batch_%d.bin' % i) for i in xrange(1, 6)]
for f in filenames:
if not tf.gfile.Exists(f):
raise ValueError('Failed to find file: ' + f)
# Create a queue that produces the filenames to read.
filename_queue = tf.train.string_input_producer(filenames)
# Read examples from files in the filename queue.
read_input = read_cifar10(filename_queue)
reshaped_image = tf.cast(read_input.uint8image, tf.float32)
height = IMAGE_SIZE
width = IMAGE_SIZE
# Image processing for training the network. Note the many random
# distortions applied to the image.
# Randomly crop a [height, width] section of the image.
distorted_image = tf.random_crop(reshaped_image, [height, width, 3])
# Randomly flip the image horizontally.
distorted_image = tf.image.random_flip_left_right(distorted_image)
# Because these operations are not commutative, consider randomizing
# the order their operation.
distorted_image = tf.image.random_brightness(distorted_image, max_delta=63)
distorted_image = tf.image.random_contrast(distorted_image, lower=0.2, upper=1.8)
# Subtract off the mean and divide by the variance of the pixels.
float_image = tf.image.per_image_whitening(distorted_image)
# Ensure that the random shuffling has good mixing properties.
min_fraction_of_examples_in_queue = 0.4
min_queue_examples = int(NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN *
min_fraction_of_examples_in_queue)
print('Filling queue with %d CIFAR images before starting to train.'
'This will take a few minutes.' % min_queue_examples)
# Generate a batch of images and labels by building up a queue of examples.
return _generate_image_and_label_batch(float_image, read_input.label,
min_queue_examples, batch_size,
shuffle=True)
are the images being distorted actually added to the pool of original images?
It depends on the definition of the pool. In tensorflow, you have ops which are basic objects in your network graph. here, data production is an op itself. Thus you do not have a finite set of training samples, instead you have a potentialy infinite set of samples generated from the training set.
or are the distorted images used instead of the originals?
As you can see from the source you included - sample is taken from the training batch, then it is randomly transformed, thus there is very small probability of using unaltered image (especially that cropping is used, which always modifies).
how many distorted images are being produced? (i.e. what augmentation factor was defined?)
There is no such thing, this is never ending process. Think about this in terms of random access to possibly infinite source of data, as this is what is efficiently happening here. Every single batch can be different from the previous one.