WINDOW_Y = 80
# load test data
print("Loading images for test...")
x_test, x_test_files = file_handler.load_images_for_keras(
my_paths.VALIDATION_PATH,
"png",
100000,
WINDOW_X,
WINDOW_Y,
num_channels=3,
scale=True)
x_test = x_test / 255
print("Loading dictionary for y_train...")
my_logger = r_logger.R_logger(os.path.join(my_paths.INFO_PATH, "data.csv"))
my_y_values = my_logger.load_dictionary()
print("Get y values...")
y_test = image_handler.get_y_value(x_test_files, my_y_values,
my_paths.human_labels)
#create y_train and y_test values
print('Number of images in x_test', x_test.shape[0])
print('Number of images in y_test', y_test.shape[0])
# load the trained neural network
from keras.models import load_model
model_name = "path_1_richardson_X1.h5"
my_model = load_model(model_name)
#dimensions of input image
WINDOW_X = 80
WINDOW_Y = 80
from keras import backend as K
print(K.tensorflow_backend._get_available_gpus())
import matplotlib.pyplot as plt
max_images = 20000
input_shape = (WINDOW_Y, WINDOW_X, 3)
print("Loading dictionary for y_train...")
my_logger = r_logger.R_logger(os.path.join(my_paths.INFO_PATH, "data.csv"))
my_y_values = my_logger.load_dictionary()
print("Loading images for train...")
x_train, x_train_files = file_handler.load_images_for_keras(
my_paths.TRAIN_PATH, "png", max_images, WINDOW_X, WINDOW_Y, num_channels=3)
x_train = x_train / 255
print("Loading images for test...")
x_test, x_test_files = file_handler.load_images_for_keras(
my_paths.VALIDATION_PATH,
"png",
max_images,
WINDOW_X,
WINDOW_Y,
num_channels=3)
boxes = load_data(richardson_path.META_FILE, richardson_path.META_PATH)
img_list = []
for label_name in richardson_path.human_labels:
rows_human = boxes.loc[boxes.LabelName == label_name]
#create a list
for row in rows_human.ImageID:
if row not in img_list: # make sure you only have unique entries!
img_list.append(row)
# select rows with ImageID
#img_id = "00a159a661a2f5aa"
from Richardson_Logger import r_logger
my_logger = r_logger.R_logger(richardson_path.INFO_PATH + '/' + "data.csv")
my_logger.clear()
my_logger.write_line("tvt,flag_person,imgid\n")
total_collection = 0 # keep track of how many images are in your dataset
for img_id in img_list:
#img_id = img_list[2]
#remove extension
img_id = img_id.split('.')[0]
select_rows = boxes.loc[boxes.ImageID == img_id]
if (len(select_rows.index) == 0):
continue
#print(select_rows)
lineType)
'''
#save image with bounding boxes to output folder
file_handler.save_image('obj_detected' + str(img_index) +'.png', path = my_paths.OBJ_TEST_RESULTS, image_data = my_image, flag_png = True, remove_color = False)
return
if os.path.exists(my_paths.OBJ_TEST_RESULTS) == False:
os.mkdir(my_paths.OBJ_TEST_RESULTS)
my_model.summary()
meta_data = data.get('_via_img_metadata')
files = file_handler.get_file_list(my_paths.TTC_PATH)
my_logger = r_logger.R_logger(os.path.join("richardson_info_files", "obj_mAP_values.csv"))
my_logger.clear()
my_logger.write_line("Detector threshold: " + str(detector_threshold) + "\n")
my_logger.write_line("Model Name: " + model_name + "\n")
my_logger.write_line("imageid, boundingbox_id, IOU, sourceimage\n")
img_index=0
for file_to_scan in files:
#get ground truth bounding box from JSON
true_bounding_boxes = get_bounding_box(meta_data, file_to_scan)
my_image = file_handler.load_image(file_to_scan, path=my_paths.TTC_PATH, flag_bw_load=False)[0]
process_image(my_image, img_index, my_logger, true_bounding_boxes)
img_index = img_index + 1