Sign Detection
--Started 1/28/2020 -
- -Keras Demo
-Pyenv
-
-As of now, Tensorflow requires python 3.7, and the current python
-version is 3.8. To provide python 3.7 to this project only, I used
-pyenv (https://github.com/pyenv/pyenv). Since I'm using Arch
-Linux, I used pacman to install pyenv, sudo pacman -S
- pyenv. Then I added the following lines to my bash_profile:
-
export PYENV_ROOT="$HOME/.pyenv" -export PATH="$PYENV_ROOT/bin:$PATH" - -if command -v pyenv 1>/dev/null 2>&1; then - eval "$(pyenv init -)" -fi --
-Then I restarted my shell and went to my project directory and ran
-pyenv local 3.7.6. Finally, to install python 3.7.6: pyenv install 3.7.6.
-
-In this directory I installed packages, via pip, for this
-version. For example, sudo pip install matplotlib. Refer to the
-github link above for more information.
-
Emacs Pyenv Mode
-
-Emacs has the pyenv-mode-auto package which, when installed,
-will automatically detect the project's python version (via the
-.python-version file generated with pyenv local 3.7.6
-above). This means that we can run org-mode python code blocks
-will use the python version detected, instead of the system python
-version (which is currently python 3.8).
-
Data Preperation
-Downloading Images from AWS S3 Bucket
-if [[ ! -d "images" ]]; then - wget https://s3.us-east-2.amazonaws.com/naturalimages02/images.tar.gz - tar -xzf images.tar.gz - rm -f images.tar.gz* -fi; -echo "DONE!" --
Common Variables
-
-common.py
-
trainPath = '../images/train/' -testPath = '../images/test/' -classCount = 8 -imageSize = 224 --
Preparing Training Data
-
-batch.py
-
from keras.preprocessing.image import ImageDataGenerator -import numpy as np -import common - -size = 16 - -trainingBatchGenerator = ImageDataGenerator( - validation_split = 0.3, - shear_range = 0.2, - zoom_range = 0.2, - horizontal_flip = True) - -trainingBatchIterator = trainingBatchGenerator.flow_from_directory( - directory = common.trainPath, - target_size = (common.imageSize, common.imageSize), - batch_size = size, - class_mode = 'categorical', - color_mode = 'rgb', - shuffle = True) - -x, y = trainingBatchIterator.next() -sampleSize = trainingBatchIterator.n --
Let's See the Batch
-import matplotlib.pyplot as pyplot -import numpy - -import batch - -figure = pyplot.figure() -columns = 4 -rows = 4 - -for i in range(1, columns*rows+1): - randomImageIndex = numpy.random.randint(batch.size) - randomImage = batch.x[randomImageIndex].astype(numpy.int) - figure.add_subplot(rows, columns, i) - pyplot.imshow(randomImage) - -pyplot.savefig(figureFile) -return figureFile --
-
Model Creation from VGG16
-
-model.py
-
import keras -from keras.models import Model, load_model -from keras.layers import Activation, Dropout, Flatten, Dense -from keras.preprocessing.image import ImageDataGenerator -from keras.applications.vgg16 import VGG16 - -import common - -baseModel = VGG16( - weights = 'imagenet', - include_top = False, - input_shape = (common.imageSize, common.imageSize, 3)) - -for layer in baseModel.layers: - layer.trainable = False - -model = keras.models.Sequential() -model.add(baseModel) - -model.add(Flatten()) -model.add(Dense(1024, activation='relu')) -model.add(Dense(1024, activation='relu')) -model.add(Dense(common.classCount, activation='softmax')) - -if __name__ == '__main__': - print(model.summary()) --
- Model: "sequential_1" - _________________________________________________________________ - Layer (type) Output Shape Param # - ================================================================= - vgg16 (Model) (None, 7, 7, 512) 14714688 - _________________________________________________________________ - flatten_1 (Flatten) (None, 25088) 0 - _________________________________________________________________ - dense_1 (Dense) (None, 1024) 25691136 - _________________________________________________________________ - dense_2 (Dense) (None, 1024) 1049600 - _________________________________________________________________ - dense_3 (Dense) (None, 8) 8200 - ================================================================= - Total params: 41,463,624 - Trainable params: 26,748,936 - Non-trainable params: 14,714,688 - _________________________________________________________________ - None --
Training
-
-training.py
-
from keras.optimizers import SGD - -import batch -from model import model - -model.compile( - loss = 'categorical_crossentropy', - optimizer = SGD(lr=1e-3), - metrics = ['accuracy']) - -# # Start the training process -# model.fit(x_train, y_train, validation_split=0.30, size=32, epochs=50, verbose=2) - -# # #save the model -# model.save('catdog.h5') - -history = model.fit_generator( - batch.trainingBatchIterator, - steps_per_epoch = batch.sampleSize/batch.size, - epochs = 10) - -model.save('fine_tune.h5') --
-723bc9aae56987359e8608626ea62441 --
Summarize History
-import matplotlib.pyplot as pyplot -import training - -pyplot.plot(training.history.history['loss']) -pyplot.title('loss') -pyplot.ylabel('loss') -pyplot.xlabel('epoch') -pyplot.legend(['loss'], loc='upper left') -pyplot.show() --