def rgbtest():
raws, exps = reader.get_images()
print(exps.keys())
# rg = color.gray2rgb(adhist)
# red_multiplier = [1, 0, 0]
# yellow_multiplier = [1, 1, 0]
# rg = red_multiplier * rg
r = exps['04']
g = exps['05']
b = exps['06']
zipped = np.dstack((r, g, b))
io.imsave(OUTDIR + "/color.png", zipped)
# r = exps['08']
# g = exps['06']
# b = exps['05']
b2 = 1-(2*exps['05'])
io.imsave(OUTDIR + "/inv.png", b2)
io.imsave(OUTDIR + "/color2.png",
np.dstack((exps['08'], exps['05'], exps['06'])))
io.imsave(OUTDIR + "/color3.png",
np.dstack((exps['04'], exps['05'], exps['08'])))
io.imsave(OUTDIR + "/color4.png",
np.dstack((exps['08'], exps['05'], exps['04'])))
io.imsave(OUTDIR + "/color6.png",
np.dstack((exps['04'], exps['08'], b2)))
io.imsave(OUTDIR + "/true_color.png",
np.dstack((exps['04'], exps['03'], exps['02'])))
io.imsave(OUTDIR + "/true_color_raw.png",
np.dstack((raws['04'], raws['03'], raws['02'])))
def train(pos, neg):
raws, exps = reader.get_images()
# flat_raws = [ndarray.flatten(raw) for raw in raws.values()]
# print(flat_raws[0].shape)
# flattened = np.dstack(flat_raws)[0]
# print(flattened.shape, flattened[0])
grid = np.dstack(raws.values())
print(grid.shape)
X = []
y = []
for p in pos:
X.append(grid[p])
y.append(1)
for p in neg:
X.append(grid[p])
y.append(0)
classifier = KNeighborsClassifier(1)
classifier = RandomForestClassifier()
classifier.fit(X, y)
testpoints = list(coords("""
553 298
430 610
500 300
500 240
"""))
testX = [grid[p] for p in testpoints]
print(classifier.predict(testX))
pred_image = np.zeros((800,800))
for i in range(800):
print(i)
# for j in range(800):
pred_image[i] = classifier.predict(grid[i])
io.imsave(OUTDIR + '/pred.png', pred_image)
orig = exps['08']
r = copy.copy(orig)
g = copy.copy(orig)
b = copy.copy(orig)
b = b + pred_image
b = np.vectorize(lambda x: min(x, 1.0))(b)
overlay = np.dstack((r, g, b))
io.imsave(OUTDIR + '/pred_overlay.png', overlay)
def show_sample(pos, neg):
raws, exps = reader.get_images()
orig = exps['08']
r = copy.copy(orig)
g = copy.copy(orig)
b = copy.copy(orig)
for p in pos:
r[p] = 1
for p in neg:
g[p] = 1
zipped = np.dstack((r, g, b))
io.imsave(OUTDIR + '/sample.png', zipped)
def make_grid():
raws, exps = reader.get_images()
orig = exps['08']
r = copy.copy(orig)
g = copy.copy(orig)
b = copy.copy(orig)
for i in range(800):
for j in range(800):
if i % 500 == 0 or j % 500 == 0:
b[i, j] = 1
elif i % 100 == 0 or j % 100 == 0:
r[i, j] = 1
elif i % 10 == 0 and j % 10 == 0:
g[i, j] = 1
#b = np.vectorize(lambda x: min(x, 1.0))(b)
zipped = np.dstack((r, g, b))
io.imsave(OUTDIR + '/s_grid.png', zipped)
# Import libraries import numpy as np import matplotlib.pyplot as plt import tensorflow as tf import keras from keras.models import Sequential from keras.layers import Input, Dense, Dropout, Flatten import pydot import graphviz # GET DATA from reader import get_images (x_train, y_train), (x_test, y_test) = get_images() num_classes = 10 y_train = keras.utils.to_categorical(y_train, num_classes) y_test = keras.utils.to_categorical(y_test, num_classes) # DEFINE PARAMETERS nInput = [10] nHidden = [200] lossFunct = ["categorical_crossentropy"] opt = ["adam"] metric = ["accuracy"] batchSize = [150] epoch = [10] # DEFINE DATA STRUCTURE TO HOLD OPTIMAL COMBINATION bestParams = [None, None, None, None, None, None, None] bestScenario = (bestParams, 0) #COUNT TOTAL NUMBER OF ITERATIONS
return red_points, green_points
def overlay(rps, gps, b_image, exp, fn):
# raws, exps = reader.get_images()
r = copy.copy(exp)
g = copy.copy(exp)
b = copy.copy(exp)
for p in rps:
r[p] = 1
for p in gps:
g[p] = 1
b = b + b_image
b = np.vectorize(lambda x: min(x, 1.0))(b)
overlay = np.dstack((r, g, b))
io.imsave(OUTDIR + '/' + fn, overlay)
pos, neg = get_sample("out/train8.png")
raws, exps = reader.get_images(3500, 3900, 800, 1200)
# show_sample(pos, neg)
pred_image = train(pos, neg, raws)
overlay(pos, neg, 0.5 * pred_image, exps['05'], 'pred_and_train.png')
overlay([], [], 0.5 * pred_image, exps['08'], 'pred_overlay.png')
