def __init__(self,
description: str = "Dataset Created Using annotation_utils",
url: str = "https://github.com/cm107/annotation_utils",
version: str = "1.0",
year: str = get_present_year(),
contributor: str = get_username(),
date_created: str = get_present_time_Ymd()):
super().__init__()
self.description = description
self.url = url
self.version = version
self.year = year
self.contributor = contributor
self.date_created = date_created
def from_scratch(self,
description: str,
url: str,
version: str = '1.0') -> COCO_Field_Buffer:
info = COCO_Info(description=description,
url=url,
version=version,
year=get_present_year(),
contributor=get_username(),
date_created=get_present_time_Ymd())
licenses = COCO_License_Handler()
images = COCO_Image_Handler()
annotations = COCO_Annotation_Handler()
categories = COCO_Category_Handler()
return COCO_Field_Buffer(info=info,
licenses=licenses,
images=images,
annotations=annotations,
categories=categories)
def process(self, save_image_to_dir: bool = True):
object_name = np.array([item.name for item in self.coco_category_list])
json_path_list = self.get_all_object_json_files()
for i, json_path in enumerate(json_path_list):
ann_dict = json.load(open(json_path, 'r'))
camera_params = self.camera_params_from_dict(ann_dict=ann_dict)
ann_object_list = ann_dict['objects']
image_location = os.path.abspath(json_path[:-5] + '.png')
coco_image = CocoImage(license=1,
file_name=os.path.basename(image_location),
coco_url=image_location,
height=camera_params.resx,
width=camera_params.resy,
date_captured=get_present_time_Ymd(),
flickr_url=None,
id=i)
self.coco_image_list.append(coco_image)
ndds_ann_object_list = []
for ann_object in ann_object_list:
ndds_ann_object = NDDS_Annotation_Object.from_dict(ann_object)
if any(ele in ndds_ann_object.class_name
for ele in object_name):
mask = np.array([
ele in ndds_ann_object.class_name
for ele in object_name
])
object_name_single = object_name[mask][0]
category_id = [
category.id for category in self.coco_category_list
if category.name == object_name_single
][0]
object_index = ndds_ann_object.class_name.replace(
object_name_single, '')
color_instance_rgb = self.segmentation_id_to_color(
ndds_ann_object)
# get segmentation and bbox
instance_img = cv2.imread(
coco_image.coco_url.replace('.png', '.is.png'))
seg = self.get_instance_segmentation(
img=instance_img,
lower_thres=(int(color_instance_rgb[0] - 1),
int(color_instance_rgb[1] - 1),
int(color_instance_rgb[2] - 1)),
upper_thresh=(int(color_instance_rgb[0] + 1),
int(color_instance_rgb[1] + 1),
int(color_instance_rgb[2] + 1)))
if len(seg) == 0:
print(
"image segmentation not found, please check color code"
)
continue
outer_bbox = seg.to_bbox()
if outer_bbox.area() < 10:
print("object too small to be considered")
continue
keypoints, keypoints_3d = self.get_keypoints(
ann_object_list=ann_object_list,
object_name_single=object_name_single,
object_index=object_index)
coco_annotation = CocoAnnotation(
bbox=outer_bbox.to_list(output_format='pminsize'),
image_id=i,
category_id=category_id,
is_crowd=0,
id=len(self.coco_annotation_list) + 1,
keypoints=keypoints,
keypoints_3d=keypoints_3d,
segmentation=seg.to_list(),
area=outer_bbox.area(),
orientation_xyzw=ndds_ann_object.quaternion_xyzw.
to_dict(),
camera_params=camera_params.to_dict_fct(),
num_keypoints=int(len(keypoints) / 3))
self.coco_annotation_list.append(coco_annotation)
if save_image_to_dir:
self.save_images()
return CocoDataset(info=self.coco_info,
licenses=self.coco_license_list,
images=self.coco_image_list,
annotations=self.coco_annotation_list,
categories=self.coco_category_list), self.save_path