def concentrated_labels(ilp_path, labels):
labels = float(labels)
#create copy project
ilp_copy_path = create_copy(ilp_path)
#work on copy file
manipulate_me = ILP(ilp_copy_path, "/home/stamylew/delme")
print
print "amount of label blocks:", manipulate_me._label_block_count(0)
blocks, block_slices = manipulate_me.get_labels(0) # 0 selects the first data set imported in the ilp
random_block_no = np.random.randint(0, len(blocks))
print
print "random_block_no", random_block_no
random_block = blocks[random_block_no]
print
print "random block shape", random_block.shape
random_slice_no = np.random.randint(0,random_block.shape[2])
smaller_slice_no = random_slice_no
bigger_slice_no = random_slice_no+1
print
print "random_slice_no", random_slice_no
random_slice = random_block[:,:,random_slice_no]
label_number = float(get_number_of_labels(random_slice))
while label_number < labels:
if bigger_slice_no < random_block.shape[0]-1:
bigger_slice_no = bigger_slice_no +1
else:
smaller_slice_no = smaller_slice_no-2
random_slice = random_block[:,:,smaller_slice_no:bigger_slice_no]
print "random slice", random_slice.shape
label_number = get_number_of_labels(random_slice)
print "label number", label_number
if label_number > labels:
break
concentrated_data = limit_label(random_slice, labels, label_number)
new_label_number = get_number_of_labels(concentrated_data)
print "new_label number", new_label_number
#create new blocks
new_blocks = []
for block in blocks:
new_block = np.zeros(block.shape, dtype=np.uint)
new_blocks.append(new_block)
# replace random block with labeled data
new_labeled_block = new_blocks[random_block_no]
new_labeled_block[:,:,smaller_slice_no:bigger_slice_no] = concentrated_data
assert get_number_of_labels(new_labeled_block) == labels
# replace old blocks with new blocks
manipulate_me.replace_labels(0, new_blocks, block_slices, delete_old_blocks=True)
print
print "new_label_number", get_number_of_labels(new_block)
return manipulate_me.project_filename
def reduce_labels_in_ilp(ilpfile, labels):
""" Reduces amount of labeled pixels to given value for labels
:param: ilpfile: path to ilastik project
:param: labels : amount labeled pixels
"""
#create copy project
ilp_copy = create_copy(ilpfile)
#work on copy file
manipulate_me = ILP(ilp_copy, "/home/stamylew/delme")
print
print "amount of label blocks:", manipulate_me._label_block_count(0)
blocks, block_slices = manipulate_me.get_labels(0) # 0 selects the first data set imported in the ilp
#save_h5(blocks, "/home/stamylew/delme/blocks.h5", "data", None)
#gather amount of labeled pixels
nolb = []
all_blocks = []
for block in blocks:
all_blocks.append(block)
nolib = get_number_of_labels(block)
nolb.append(nolib)
noal = float(np.sum(nolb))
print
print "amount of all labels:", noal
labels = float(labels)
#limit amount of labeled pixels
assert labels < noal
percentage = labels/noal
print
print "percentage of labels:", percentage
new_blocks = []
for block in blocks:
new_block = filter_all_labels(block, percentage)
new_blocks.append(new_block)
#save_h5(new_blocks, "/home/stamylew/delme/new_blocks.h5", "data", None)
#check new blocks
nolnb = []
for block in new_blocks:
nolinb = get_number_of_labels(block)
nolnb.append(nolinb)
# print "unique elements of new blocks:", np.unique(new_blocks)
nonl = float(np.sum(nolnb))
print
print "reduced amount of labels:", nonl
# assert labels == nonl
manipulate_me.replace_labels(0, new_blocks, block_slices, delete_old_blocks=True)
return manipulate_me.project_filename
def __init__(self, gt, predict, threshold = 0.5):
self.gt = gt
self.adjusted_predict = adjust_predict(predict)
self.relevant_data = exclude_ignore_label(gt, self.adjusted_predict)
self.nlables = get_number_of_labels(gt)
self.apr = apr(self.relevant_data[0], self.relevant_data[1])
self.auc_score = roc_auc(self.relevant_data[0], self.relevant_data[1])
self.quality = np.array((self.apr[0], self.apr[1], self.apr[2], self.auc_score))
self.roc_curve = draw_roc_curve(self.relevant_data[0], self.relevant_data[1])
def check_ilp_labels(ilp_path):
ilp = ILP(ilp_path, "/home/stamylew/delme")
blocks, blockslices = ilp.get_labels(0)
list = []
for block in blocks:
noilb = get_number_of_labels(block)
list.append(noilb)
al = list[0]
for i in range(1, len(list)):
al += list[i]
print "reduced amount of labels", al
#create new blocks
new_blocks = []
for block in blocks:
new_block = np.zeros(block.shape, dtype=np.uint)
new_blocks.append(new_block)
# replace random block with labeled data
new_labeled_block = new_blocks[random_block_no]
new_labeled_block[:,:,smaller_slice_no:bigger_slice_no] = concentrated_data
assert get_number_of_labels(new_labeled_block) == labels
# replace old blocks with new blocks
manipulate_me.replace_labels(0, new_blocks, block_slices, delete_old_blocks=True)
print
print "new_label_number", get_number_of_labels(new_block)
return manipulate_me.project_filename
if __name__ == '__main__':
trimap_path = "/home/stamylew/volumes/groundtruth/trimaps/500p_cube2_trimap_t_05.h5"
#trimap_path = "/home/stamylew/volumes/groundtruth/trimaps/100p_cube1_trimap_t_05.h5"
trimap_data = read_h5(trimap_path, "data")
print "trimap labeled pixel amount", get_number_of_labels(trimap_data)
ilp_path = "/home/stamylew/ilastik_projects/100p_cube1.ilp"
#ilp_path = "/home/stamylew/ilastik_projects/100p_cube1.ilp"
# mod_ilp = modify_labels_of_ilp(ilp_path, 3000)
# check_ilp_labels(mod_ilp)
concentrated_data = concentrated_labels(ilp_path, 1000)
# save_h5(concentrated_data, "/home/stamylew/delme/concentrated_ilp_labels.h5", "data", None)
print "done"
