ImageAnnotationsIO provides methods for loading and saving image annotations using common formats, like the CVAT XML annotation format, or the LabelMe XML format.
There is currently support for loading and saving:
- Bounding box annotations (for object detection).
- Image annotations (for image classification).
- Oriented bounding box annotations (for object detection).
- Polygon annotations (for object detection).
The following example loads and saves CVAT XML using decimal floating-point representation of coordinates:
using DecFP
using ImageAnnotationsIO
cvat_xml_serializer = CVATXMLSerializer{Dec128}()
dataset = load(input_path, cvat_xml_serializer)
save(output_path, dataset, cvat_xml_serializer)The following example loads a dataset in the form of directories containing LabelMe XML format files, and saves it in the form of a single CVAT XML file.
using Downloads
using ImageAnnotations
using ImageAnnotationsIO
using ImageIO
base_path = mktempdir()
# Download LabelMe XML dummy data
input_path = joinpath(base_path, "example_folder")
mkdir(input_path)
Downloads.download("https://raw.githubusercontent.com/CSAILVision/LabelMeAnnotationTool/master/Annotations/example_folder/img1.xml", joinpath(input_path, "img1.xml"))
Downloads.download("https://raw.githubusercontent.com/CSAILVision/LabelMeAnnotationTool/master/Images/example_folder/img1.jpg", joinpath(input_path, "img1.jpg"))
output_path = joinpath(base_path, "example_folder_cvat.xml")
labelme_xml_serializer = LabelMeXMLSerializer{Float64}() # Serialize/Deserialize LabelMe XML using Float64 coordinate type
cvat_xml_serializer = CVATXMLSerializer{Float64}() # Serialize/Deserialize CVAT XML using Float64 coordinate type
data_set = load_dataset_dir(input_path, labelme_xml_serializer; base_path = base_path, image_base_path = base_path)
save(output_path, data_set, cvat_xml_serializer)Please note that the example will issue a few warnings related to the current lack of support for object segmentation annotations.