Homec4science
Diffusion solarPV (Dan_master_thesis) Browse

solarPV_identification/CNN/segnet_main.py
e8b57b448f0cDan_master_thesis

Last Change
Raw File

segnet_main.py
View Options

import os
import tensorflow
import numpy as np
import scipy
import data_loaders
import datetime
import sys
# import cv2
from scipy import ndimage
#from matplotlib import pyplot as plt
from PIL import Image
from keras.models import *
from keras.layers import Reshape
from keras.layers.core import *
from keras.layers import *
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import *
from keras.layers.advanced_activations import LeakyReLU
from keras.optimizers import Adam, SGD
from keras.utils import np_utils
from keras import backend as K
def weighted_categorical_crossentropy(weights):
"""
A weighted version of keras.objectives.categorical_crossentropy
Variables:
weights: numpy array of shape (C,) where C is the number of classes
Usage:
weights = np.array([0.5,2,10]) # Class one at 0.5, class 2 twice the normal weights, class 3 10x.
loss = weighted_categorical_crossentropy(weights)
model.compile(loss=loss,optimizer='adam')
"""
weights = K.variable(weights)
def loss(y_true, y_pred):
# scale predictions so that the class probas of each sample sum to 1
y_pred /= K.sum(y_pred, axis=-1, keepdims=True)
# clip to prevent NaN's and Inf's
y_pred = K.clip(y_pred, K.epsilon(), 1 - K.epsilon())
# calc
loss = y_true * K.log(y_pred) * weights
loss = -K.sum(loss, -1)
return loss
return loss
default_w = [1, 1]
def segnet(nClasses, optimizer=None, input_height=360, input_width=480, weights=default_w):
kernel = 3 # Size of the kernel
first_conv_windows = 16 # Will determine complexity of the model
# Original is 64 yields 29.000.000 parameters
# 4 yields 21.785 parameters
pad = 3 # so that image dimensions are at least 2*2*2*2*n shape (4 maxpooling...)
pool_size = 2
model = Sequential()
# encoder
model.add(ZeroPadding2D(padding=3, input_shape=(input_height, input_width, 3)))
model.add(Conv2D(first_conv_windows, kernel_size=kernel, padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D())
model.add(Conv2D(first_conv_windows * 2, kernel_size=kernel, padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D())
model.add(Conv2D(first_conv_windows * 4, kernel_size=kernel, padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D())
model.add(Conv2D(first_conv_windows * 8, kernel_size=kernel, padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
# Decoder
model.add(Conv2D(first_conv_windows * 8, kernel_size=kernel, padding='same'))
model.add(BatchNormalization())
model.add(UpSampling2D(size=(pool_size, pool_size)))
model.add(Conv2D(first_conv_windows * 4, kernel_size=kernel, padding='same'))
model.add(BatchNormalization())
model.add(UpSampling2D(size=(pool_size, pool_size)))
model.add(Conv2D(first_conv_windows * 2, kernel_size=kernel, padding='same'))
model.add(BatchNormalization())
model.add(UpSampling2D(size=(pool_size, pool_size)))
model.add(Conv2D(first_conv_windows, kernel_size=kernel, padding='same'))
model.add(BatchNormalization())
model.add(Conv2D(nClasses, kernel_size=1, padding='same'))
model.add(Cropping2D(cropping=pad))
model.add(Reshape((nClasses, input_height * input_width)))
model.add(Permute((2, 1)))
model.add(Activation('softmax'))
if not optimizer is None:
model.compile(loss=weighted_categorical_crossentropy(weights), optimizer=optimizer, metrics=['accuracy'])
return model
def run(rotations, lights, dice, weight1, weight2, epochs):
original_images, labeled_images = data_loaders.load_images(lights=lights, rotations=rotations)
X_train = np.array(original_images)
Y_train_flat = np.array(labeled_images).reshape(len(labeled_images), -1)
# One dimension for each channel (one channel PV, one channel non-PV)
Y_train_2d = np.stack((np.where(Y_train_flat == 0, 1, 0), Y_train_flat), axis=-1).reshape(len(X_train), -1, 2)
inp_shape = X_train[0].shape
out_shape_flat = Y_train_flat[0].shape
out_shape_2d = Y_train_2d[0].shape
model_u = segnet(2, optimizer='adam', input_width=250, input_height=250, weights=[weight1, weight2])
model_u.summary()
model_name = str(datetime.date.today()) + "segnet_dice_{:}_{:}_epochs:_{:}".format(dice, (weight1, weight2), epochs)
print(model_name)
model_u.fit(X_train, Y_train_2d, validation_split=0.1, epochs=epochs)
# Save the weights
model_u.save('/home/mesguerr/models/segnet{:}.h5'.format(model_name))
# Save the model architecture
if __name__ == "__main__":
rotations = int(sys.argv[1])
lights = int(sys.argv[2])
dice = int(sys.argv[3])
weight1 = float(sys.argv[4])
weight2 = float(sys.argv[5])
epochs = int(sys.argv[6])
run(rotations, lights, dice, weight1, weight2, epochs)
c4science · Help