def create_data(folder, sameSize, maxsize):
count_classes = 0
min_data = -1
data = []
label = []
print(str(len(os.listdir(folder))) + " classes found")
for classe in os.listdir(folder):
print(classe)
if (classe == ".gitkeep"):
continue
print("loading classe : " + fm.make_path(folder, classe, "*.jpg") +
"\n")
data_current = load_pictures(fm.make_path(folder, classe, "*.jpg"),
maxsize)
label_current = [count_classes for i in range(len(data_current))]
data.append(data_current)
label.append(label_current)
count_classes = count_classes + 1
if min_data == -1 or min_data > len(data_current):
min_data = len(data_current)
dataToReturn = []
labelToReturn = []
if sameSize:
for i in range(len(data)):
dataToReturn = dataToReturn + data[i][:min_data]
labelToReturn = labelToReturn + label[i][:min_data]
return (np.array(dataToReturn), np.array(labelToReturn))
def write_image(array, directory, file_name):
fm.clean_folder(fm.make_path('polyps', 'output', 'data'))
index = 0
img_store = []
for image in array:
img = Image.new('RGB', config['image_size'], "white")
for i in range(config['image_max']):
for j in range(config['image_max']):
img.putpixel((i, j), image[i][j])
n = (file_name[index].split(config['path_sep']))[-1]
name = '{}'.format(n)
img.save(fm.make_path(directory, name))
img_store.append(np.array(img))
index = index + 1
return img_store
def createMask(folderIn, masks):
store_im = []
cpt = 0
for filename in [make_path(folderIn, name) for name in list_dir(folderIn)]:
picture = pixelsVerification(load_image(filename))
for i in range(len(masks)):
store_im.append(cv2.inRange(picture, masks[i], masks[i]) / 255)
print("\rPicture - " + str(cpt), end='')
cpt += 1
return store_im
str(round(tn, 3)) + " FN: " + str(round(fn, 3)))
if __name__ == "__main__":
data_augmentation.execute()
batch_size = config['batch_size']
#model = model.u_net(IMG_SIZE = config['image_dimension']) #what does the Adam optimizer do
model = model.u_net_batch_norm_upc()
model.compile(optimizer=Adam(lr=1e-4),
loss=config['loss'],
metrics=config['metrics'])
img_train = np.array(
load_transform_pictures(
fm.make_path('polyps', 'training', 'data', '*.jpg')))
img_val = np.array(
load_transform_pictures(
fm.make_path('polyps', 'validation', 'data', '*.jpg')))
img_test = np.array(
load_transform_pictures(fm.make_path('polyps', 'test', 'data',
'*.jpg')))
test_name = np.array(
load_file_name(fm.make_path('polyps', 'test', 'data', '*.jpg')))
output = fm.make_path('polyps', 'output', 'classification')
mask_train = np.array(data_transformation.create_binary_masks(
path=fm.make_path('polyps', 'training', 'label')),
dtype="float32")
# Display the first images in data
classCNN.plot_data(train_dataSet, test_dataSet)
# %% Model Creation
model1 = classCNN.createModel(train_data.shape[1:], nClasses,
config['model'])
model1.compile(optimizer=config['compile']['optimizer'],
loss=config['compile']['loss'],
metrics=config['compile']['metrics'])
model1.summary()
# Open the file
with open(
fm.make_path(config['folderModel'], config['nameModel'],
config['nameModel'] + '.txt'), 'w') as fh:
# Pass the file handle in as a lambda function to make it callable
model1.summary(print_fn=lambda x: fh.write(x + '\n'))
# %% fit the program
print("Hit a key to train the model ...")
input()
callBacks1 = callbacks.EarlyStopping(
monitor=config['callbacks1']['monitor'],
min_delta=config['callbacks1']['min_delta'],
patience=config['callbacks1']['patience'],
verbose=config['callbacks1']['verbose'],
mode=config['callbacks1']['mode'],
baseline=config['callbacks1']['baseline'],
def plot_fullInformation(predictPerCent, history, predict_succes,
predict_fails, path, name, analysis):
if (len(analysis) < 4):
raise ValueError('plot_fullInformation : no enough analysis (<4)')
nbPicturesTested = analysis[0] + analysis[1] + analysis[2] + analysis[3]
cells = [["Class :", "Clean", "Dirty"],
["True", str(analysis[0]),
str(analysis[2])], ["False",
str(analysis[1]),
str(analysis[3])]]
fig = plt.figure(3)
fig.clear()
gs = GridSpec(2, 2)
gs.update(hspace=0.4, bottom=0.05)
fig.suptitle('Model name : ' + str(name), fontsize=18)
gs1 = GridSpecFromSubplotSpec(1, 2, subplot_spec=gs[0, :])
ax1 = plt.subplot(gs1[0])
ax1.plot(history.history['loss'], 'r-+', linewidth=1)
ax1.plot(history.history['val_loss'], 'b-+', linewidth=1)
ax1.legend(['Training Accuracy', 'Validation Accuracy'],
fontsize=10,
loc='upper right',
borderaxespad=0.)
ax1.grid(True)
ax1.set_xlabel('Epochs ')
ax1.set_title('Loss Curves', fontsize=12)
ax2 = plt.subplot(gs1[1])
ax2.plot(history.history['acc'], 'r-+', linewidth=1)
ax2.plot(history.history['val_acc'], 'b-+', linewidth=1)
ax2.grid(True)
ax2.set_xlabel('Epochs ')
ax2.set_title('Accuracy Curves', fontsize=12)
# box plot part
gs2 = GridSpecFromSubplotSpec(2, 2, subplot_spec=gs[1, 0], wspace=0.3)
# succes plot
ax3 = plt.subplot(gs2[:, 0])
bp = ax3.boxplot(predict_succes, patch_artist=True)
ax3.axis([0.75, 1.25, 0.4, 1])
ax3.set_title('Succes (' + str(round(predictPerCent, 2)) + '%)',
fontsize=12)
ax3.grid(True)
ax3.set_ylabel('Probability given by the prediction')
# fails plot
ax4 = plt.subplot(gs2[:, 1])
bp2 = ax4.boxplot(predict_fails, patch_artist=True)
ax4.axis([0.75, 1.25, 0.4, 1])
ax4.set_title('Fails (' + str(round(100 - predictPerCent, 2)) + '%)',
fontsize=12)
ax4.grid(True)
for element in ['boxes']:
plt.setp(bp[element], color='green')
plt.setp(bp2[element], color='red')
for patch in bp['boxes']:
patch.set(facecolor='honeydew')
for patch in bp2['boxes']:
patch.set(facecolor='mistyrose')
# test and table
gs3 = GridSpecFromSubplotSpec(2, 4, subplot_spec=gs[1, 1])
# print the table
ax5 = plt.subplot(gs3[0, 0:2])
ax5.axis('off')
colorCell = [['white', 'white', 'white'],
['tab:green', 'lightgreen', 'lightgreen'],
['tab:red', 'mistyrose', 'mistyrose']]
table = ax5.table(cellText=cells, loc='center', cellColours=colorCell)
ax5.set_title('Analysis of the prediction', fontsize=12)
ax5.text(0.5,
-0.05,
'(' + str(nbPicturesTested) + ' pictures)',
horizontalalignment='center',
fontsize=12)
for i in range(3):
for u in range(3):
table.get_celld()[(u, i)]._loc = 'center'
table.get_celld()[(u, i)].set_width = 2
table.get_celld()[(u, i)].set_height = 2
# Reading of the summary file
summary = ""
with open(fm.make_path(path, name, name + '.txt'), 'r') as myfile:
summary += myfile.read().replace("=", "*")
ax6 = plt.subplot(gs3[:, 2])
ax6.axis('off')
ax6.text(0, 1, summary, fontsize=5, verticalalignment='top')
ax6.set_title('Model summary', fontsize=12, horizontalalignment='left')