def combining_imageWith_fixations(filtered_experiment_class,
dir_images,
memory,
search_type):
"""Create Tripplets [Imagename (String),
EyeMovements (DataFrame), Imagepath (path)]
Argument:
filtered_experiment_class: (DataFrame) Eye Movements
dir_images: (Path) Image Directory
memory: (String) memory or search
search_type: (String) (String)
'target_not_present'
'expectedlocation'
'randomlocation'
Returns:
Tripplets (Lists)
"""
path =0
image_paths = get_img_paths(dir_images)
list_of_masks = []
for i, image in enumerate(np.unique(filtered_experiment_class["imageid"]) ):
for image_path in image_paths:
if memory ==1:
#in case of memory
if int(re.findall(r'\d+',
os.path.basename(image_path))[0]) == image:
path = image_path
elif memory==0:
if search_type == "target_not_present":
match = "L_"
elif search_type == 'expectedlocation':
match = "E_"
else:
match = "U_"
#in case of search
if int(re.findall(r'\d+',
os.path.basename(image_path))[0]) == image and match in image_path:
path = image_path
dataframe_temp =\
filtered_experiment_class[filtered_experiment_class["imageid"] == image]
if path==0:
raise ValueError("wrong image path !")
list_of_masks.append([str(i+1)+".png",
path,
dataframe_temp])
return list_of_masks
def load_images_and_masks(main_dir, height, width, br=None):
"""Load images and masks into numpy arrays.
Arguments:
main_dir: directory containing images and masks in the subdirectories
images and masks
height: target image height for resizing
width: target image width for resizing
br: number of images and masks to load
Returns:
Images and masks as a tuple of two numpy arrays.
"""
all_image_paths = get_img_paths(main_dir + "/images/")
all_mask_paths = get_img_paths(main_dir + "/masks/")
image_ids = [os.path.basename(i) for i in all_image_paths]
mask_ids = [os.path.basename(i) for i in all_mask_paths]
if br:
image_ids = image_ids[:br]
mask_ids = mask_ids[:br]
image_list = []
mask_list = []
for n, (image_id, mask_id) in tqdm(enumerate(zip(image_ids, mask_ids)),
total=len(image_ids)):
try:
image = read_image(main_dir + '/images/' + image_id, height, width)
mask = read_mask(main_dir + '/masks/' + mask_id, height, width)
except Exception as e:
print('Skipped image and mask:', n, image_id, mask_id, e)
continue
image_list.append(np.float16(image / 255))
mask_list.append(np.float16(mask / 255))
X = np.stack(image_list)
Y = np.stack(mask_list)
return X, Y
def load_images_and_masks(path, down_size=2, br=100000):
"""Load the images and preprocess for the training and prediction
input: path to the images and mask in two seperate directorys
returns: images, masks
"""
all_image_paths = get_img_paths(path + "/images/")
train_ids = [os.path.basename(i) for i in all_image_paths]
X_images = np.zeros((len(train_ids), int(
IMG_HEIGHT / down_size), int(IMG_WIDTH / down_size), 3),
dtype=np.float)
Y_images = np.zeros((len(train_ids), int(
IMG_HEIGHT / down_size), int(IMG_WIDTH / down_size), 1),
dtype=np.float)
i = 0
for n, id_ in tqdm(enumerate(train_ids), total=len(train_ids)):
image = np.zeros((IMG_HEIGHT, IMG_WIDTH, 3), dtype=np.float)
try:
img = imread(path + "/images/" + id_)
if len(img.shape) == 3:
image = img[:, :, :3]
else:
image[:, :, 0] = img[:, :]
image[:, :, 1] = img[:, :]
image[:, :, 2] = img[:, :]
image = resize(
image,
(int(IMG_HEIGHT / down_size), int(IMG_WIDTH / down_size)),
mode='constant',
preserve_range=True)
mask = imread(path + "/masks/" + id_)
mask = np.expand_dims(resize(
mask,
(int(IMG_HEIGHT / down_size), int(IMG_WIDTH / down_size)),
mode='constant',
preserve_range=True),
axis=-1)
X_images[i] = image / 255
Y_images[i] = mask / 255
i += 1
if i == br: break
except:
print("ERROR")
return X_images, Y_images
def load_data(path, down_size=16):
"""
"""
#loading data
img_height = 768
img_width = 1024
img_channels = 3
# Get train and test IDs
train_ids = [os.path.basename(i) for i in get_img_paths(path + "images")]
# Get and resize train images and masks
X_train = np.zeros((len(train_ids), int(
img_height / down_size), int(img_width / down_size), img_channels),
dtype=np.float)
Y_train = np.zeros((len(train_ids), int(
img_height / down_size), int(img_width / down_size), 1),
dtype=np.float)
i = 0
sys.stdout.flush()
for n, id_ in tqdm(enumerate(train_ids), total=len(train_ids)):
try:
img = imread(path + "images" + "/" + id_)[:, :, :img_channels]
except:
return X_train, Y_train
try:
mask = imread(path + "masks/" + id_)
except:
return X_train, Y_train
img = resize(img,
(int(img_height / down_size), int(img_width / down_size)),
mode='constant',
preserve_range=True)
mask = np.expand_dims(resize(
mask, (int(img_height / down_size), int(img_width / down_size)),
mode='constant',
preserve_range=True),
axis=-1)
#X_train[n] = img / 255
# Y_train[n] = mask / 255
X_train[n] = img / 255
Y_train[n] = mask / 255
return X_train, Y_train
def get_image_objects_memory(filtered_search_data, dir_images):
"""creating objects each pictures with different experiment types
input: filtered_data: Dataframe of the Experients,
dir_images: Directory of the images
return: objects of each image with eyemovements and pictures
"""
#get the image paths
image_paths = get_img_paths(dir_images)
original_paths = [
i for i in image_paths if "original" in i and "._O" not in i
]
memory_paths = [i for i in original_paths if "memory" in i]
list_of_exp = classify_data(filtered_memory_data)
list_of_masks = []
for u in range(len(list_of_exp)):
for i in range(len(np.unique(list_of_exp[u][0]["imageid"]))):
if list_of_exp[u][3] == "memory":
for z, image_path in enumerate(memory_paths):
if list_of_exp[u][2] in image_path:
if os.path.basename(image_path) == str(i + 1) + ".png":
path = image_path
dataframe_temp = list_of_exp[u][0][list_of_exp[u][0]["imageid"] ==
i + 1]
if dataframe_temp.shape[0]:
if list_of_exp[u][3] == "memory":
list_of_masks.append([
list_of_exp[u][0][list_of_exp[u][0]["imageid"] == i +
1], list_of_exp[u][1],
list_of_exp[u][2], list_of_exp[u][3], path,
dataframe_temp.shape[0]
])
return list_of_masks
def rotate_allimages(path, degree):
"""Rotate every images in the given directory, with the given degree
input: directory path, degree to rotate
"""
image_paths = get_img_paths(path)
# iterate through the names of contents of the folder
for image_path in image_paths:
# create the full input path and read the file
image_to_rotate = ndimage.imread(image_path)
# rotate the image
rotated = ndimage.rotate(image_to_rotate, degree)
# create full output path, 'example.jpg'
# becomes 'rotate_example.jpg', save the file to disk
#fullpath = os.path.join(outPath, 'rotated_'+image_id)
misc.imsave(image_path, rotated)
def imageWith_fixations(filtered_experiment_class, dir_images, memory,
search_type):
"""This function combinates the experiment DataFrame with Image
Argument:
DataFrame: filtered Experiment class
Directory: Image Directory
Returns:
List of connected imagepaths with fixation DataFrames
"""
image_paths = get_img_paths(dir_images)
list_of_masks = []
for i, image in enumerate(np.unique(filtered_experiment_class["imageid"])):
for image_path in image_paths:
if memory == 1:
#in case of memory
if int(re.findall(r'\d+',
os.path.basename(image_path))[0]) == image:
path = image_path
elif memory == 0:
if search_type == "target_not_present":
match = "L_"
elif search_type == 'expectedlocation':
match = "E_"
else:
match = "U_"
#in case of search
if int(re.findall(r'\d+', os.path.basename(image_path))
[0]) == image and match in image_path:
path = image_path
dataframe_temp = filtered_experiment_class[
filtered_experiment_class["imageid"] == image]
list_of_masks.append([str(i + 1) + ".png", path, dataframe_temp])
return list_of_masks
def get_image_objects(filtered_data,dir_images):
"""creating objects each pictures with different experiment types
input: filtered_data: Dataframe of the Experients,
dir_images: Directory of the images
return: objects of each image with eyemovements and pictures
"""
#get the image paths
image_paths = get_img_paths(dir_images)
original_paths = [i for i in image_paths if "original" in i and "._O" not in i]
memory_paths = [i for i in original_paths if "memory" in i]
search_paths = [i for i in original_paths if "search" in i]
list_of_exp = seperate(filtered_data)
list_of_masks = []
for u in range(len(list_of_exp)):
for i in range(len(np.unique(list_of_exp[u][0]["imageid"]) )):
if list_of_exp[u][3] == "memory":
for z, image_path in enumerate(memory_paths):
if list_of_exp[u][2] in image_path:
if os.path.basename(image_path) == str(i+1)+".png":
path = image_path
elif list_of_exp[u][3] == "search":
if list_of_exp[u][4] == 'target_not_present':
match = "L_"
else:
if list_of_exp[u][5] == 'expectedlocation':
match = "E_"
else:
match = "U_"
path_0 = [f for f in search_paths if (i+1) == int(re.findall(r'\d+',
os.path.basename(f))[0]) and match in f and list_of_exp[u][2] in f]
try:
path =path_0[0]
except:
pass
dataframe_temp = list_of_exp[u][0][list_of_exp[u][0]["imageid"] == i+1]
if dataframe_temp.shape[0]:
if list_of_exp[u][3] == "memory":
list_of_masks.append(
[list_of_exp[u][0][list_of_exp[u][0]["imageid"] == i+1],
list_of_exp[u][1],
list_of_exp[u][2],
list_of_exp[u][3],
path,
dataframe_temp.shape[0]])
elif list_of_exp[u][3] == "search":
list_of_masks.append(
[list_of_exp[u][0][(list_of_exp[u][0]["imageid"] == i+1) ],
list_of_exp[u][1],
list_of_exp[u][2],
list_of_exp[u][3],
list_of_exp[u][4],
list_of_exp[u][5],
path,
dataframe_temp.shape[0]])
return list_of_masks
'mean_pred' : mean_pred,
'max_pred': max_pred,
'min_pred' : min_pred,
'mean_true': mean_true,
'max_true': max_true,
'min_true':min_true})
else:
print('Using normal model.')
model = vgg16_enc_dec_model((img_height, img_width, img_channels),
freeze_vgg = False,
output_choice = 'sigmoid',
**{'reg':None,'dropout':0.1})
model.load_weights(weights_path)
if tr_or_val == 'tr':
image_paths = get_img_paths(path_train+"/images")
mask_paths = get_img_paths(path_train+"/masks")
else:
image_paths = get_img_paths(path_test+"/images")
mask_paths = get_img_paths(path_test+"/masks")
#TODO: If there is a problem with NaN values, one can try this:
#for n,layer in enumerate(model.layers):
# weights = layer.get_weights()
# new_weights = [np.nan_to_num(w) for w in weights]
# layer.set_weights(new_weights)
#%%
if nb_random_shows > len(image_paths):
print('Showing all images since number of random shows exceed set lengths.')
for n, (path,mask) in enumerate(zip(image_paths, mask_paths)):