def testWriteLumiImagesCsv(self):
# Test get lumi csv dataframe
# Set inputs to write_lumi_images_csv
bb_labels = add_basename_gather_df(self.bb_ann_filenames,
self.input_image_format)
output_dir = tempfile.mkdtemp()
output_ann_path = os.path.join(output_dir, "test_train.csv")
write_lumi_images_csv(
self.images,
output_dir,
self.input_image_format,
self.output_image_format,
bb_labels,
output_ann_path,
)
images = natsort.natsorted(
glob.glob(os.path.join(output_dir,
"*" + self.output_image_format)))
# Assert lumi_df is as expected and the images are written in the
# expected output_dir
assert len(images) == len(self.images)
lumi_df = pd.read_csv(output_ann_path)
assert len(lumi_df) == self.num_bboxes * self.num_images
assert list(lumi_df.columns.values) == LUMI_CSV_COLUMNS
def testGetLumiCsvDf(self):
# Test get lumi csv dataframe
# Set inputs to get_lumi_csv_df
df = add_basename_gather_df(self.bb_ann_filenames,
self.input_image_format)
images = [image.replace(os.sep, "_") for image in self.images]
lumi_df = get_lumi_csv_df(df, images, self.input_image_format)
# Assert lumi_df is as expected
assert len(lumi_df) == self.num_bboxes * self.num_images
assert list(lumi_df.columns.values) == LUMI_CSV_COLUMNS
def testAddBasenameGatherDfTxt(self):
# Test combine list of annotation txt files & add base_path col
# Set inputs to add_basename_gather_df
df = add_basename_gather_df(self.bb_ann_filenames,
self.input_image_format)
# Assert base_path column exists and is as expected
for index, row in df.iterrows():
assert row["base_path"] == (
os.path.dirname(row["image_id"]).replace(os.sep, "_") + "_" +
os.path.basename(row["image_id"]).replace(
self.input_image_format, ""))
def testGetImagePathsPerClass(self):
# Test get unique image paths per class
# Set inputs to get_image_paths_per_class
bb_ann_filenames = self.get_ann_filenames(
self.num_images,
[
[0] * self.num_bboxes,
[1] * self.num_bboxes,
self.labels,
self.labels,
[0] * self.num_bboxes,
],
)
df = add_basename_gather_df(bb_ann_filenames, self.input_image_format)
image_ids_per_class = get_image_paths_per_class(df)
# Assert image_ids_per_class are as expected
expected_key_values_length = {0: 4, 1: 3, 2: 2}
for key, value in expected_key_values_length.items():
assert value == len(image_ids_per_class[key])
def testFilterDenseAnnotation(self):
# Test filtered unique image paths per class test
# after suppressing the dense annotation of a class
# Set inputs to filter_dense_annotation
bb_ann_filenames = self.get_ann_filenames(
self.num_images,
[
[0] * self.num_bboxes,
[1] * self.num_bboxes,
self.labels,
self.labels,
[0] * self.num_bboxes,
],
)
filtered_image_ids_per_class = filter_dense_annotation(
get_image_paths_per_class(
add_basename_gather_df(bb_ann_filenames,
self.input_image_format)))
# Assert filtered_image_ids_per_class are as expected
expected_key_values_length = {1: 3, 2: 2}
for key, value in expected_key_values_length.items():
assert value == len(filtered_image_ids_per_class[key])