# Manual settings
bridge = CvBridge()
orig_size = (720, 576)
depth_orig = (558, 375)
target_size = (64, 64)
crop = ((55, 431), (107, 665))
# Create Flapnet objects
fn_struct = flapnet.Structure()
fn_losses = flapnet.LossFunction()
fn_preproc = flapnet.Preprocessing()
fn_post = flapnet.Postprocessing()
img_proc = grapof.ImageProcessing()
ts = grapof.TopicsSubscription()
geo = grapof.Geometry()
fn = flapnet.Functions(shape_img=(64, 64, 3))
def callback(image_msg):
# define global variable
global cam_disp
global cv_image
# convert image to a compatible format
cv_image = bridge.imgmsg_to_cv2(image_msg, desired_encoding='bgr8')
cv_image = cv_image[crop[0][0]:crop[0][1], crop[1][0]:crop[1][1]]
cam_disp = fn_preproc.image_preproc(cv_image, target_size)
# ROS init
rospy.init_node('camera_flap_detection', anonymous=True)
mpl.rcParams['figure.figsize'] = (12, 12)
import os
import flapnet
import numpy as np
import matplotlib as mpl
import pandas as pd
import tensorflow as tf
from tensorflow.python.keras import models
mpl.rcParams['axes.grid'] = False
mpl.rcParams['figure.figsize'] = (12, 12)
model_path = '/home/aleks/nn_ftw.hdf5'
testset_size = 0.1
fn = flapnet.Functions()
fn_struct = flapnet.Structure()
fn_losses = flapnet.LossFunction()
# Da/home/aleks/nn_results/Gtaset folders init
dataset_name = os.path.join('tissue_dataset', 'cyst_dataset')
img_dir = os.path.join(dataset_name, "train")
label_dir = os.path.join(dataset_name, "label")
df_train = pd.read_csv(os.path.join(dataset_name, 'cyst_dataset.csv'))
# Load filenames of labels and tra ining objects
x_train_filenames, x_val_filenames, y_train_filenames, y_val_filenames = fn.load_filenames(
df_train=df_train,
img_dir=img_dir,
label_dir=label_dir,
test_size=testset_size)
#---------------------------- PARAMETERS ---------------------------#
#####################################################################
img_shape = (64, 64, 3) # Input image shape
batch_size = 30 # Batch size for training (decrease in case of memory error)
epochs = 200 # Training epochs
testset_size = 0.1 # Percentage of training set data (0.10 = 10%)
dropout_rate = 0.4 # Droput rate
num_filters = 32 # Number of neurons in first layer (subsequent layers have x2 neurons)
learning_rate = 0.001 # Learning rate tuning for optimizer
adam_opt = tf.keras.optimizers.Adam(lr=learning_rate)
kfold = KFold(n_splits=6, shuffle=True, random_state=42)
# FlapNet class init
fn = flapnet.Functions(img_shape, batch_size, epochs)
fn_struct = flapnet.Structure()
fn_losses = flapnet.LossFunction()
# Dataset folders init
# dataset_name = os.path.join('tissue_dataset', 'dataset_baseline')
# img_dir = os.path.join(dataset_name, "train")
# label_dir = os.path.join(dataset_name, "labels")
# df_train = pd.read_csv(os.path.join(dataset_name,'label_map_00.csv'))
# #
dataset_name = os.path.join('tissue_dataset', 'dataset_ready_aug_02')
img_dir = os.path.join(dataset_name, "train")
label_dir = os.path.join(dataset_name, "labels")
df_train = pd.read_csv(os.path.join(dataset_name, 'ready_dataset_aug_02.csv'))
#####################################################################