ONNXRuntimeImageAnnotators

ONNXRuntimeImageAnnotators provides implementations of the ImageAnnotations.jl AbstractImageAnnotator interface using ONNXRunTime.jl.

Limitations

Support is currently limited to the YOLOv7 models.

Usage

Using the TestImages image lighthouse, bounding box annotations can be computed using a YOLOv7 model:

using DataDeps
using ImageAnnotations
using ImageTransformations
using ONNXRuntimeImageAnnotators
using TestImages

include(normpath(dirname(pathof(ONNXRuntimeImageAnnotators)), "..", "test", "onnx_models.jl"))

model_path = joinpath(datadep"WongKinYiu_yolov7_v0.1_yolov7-nms-640", "yolov7-nms-640.onnx")
model_metadata = Dict(:input_name => "images", :output_name => "output")
annotator = ONNXRuntimeObjectAnnotator{Int, Float32, BoundingBoxAnnotation{Int, Float32}}(model_path, model_metadata) # Create an ONNXRuntimeObjectAnnotator using Int label type, Float32 coordinate type, and corresponding BoundingBoxAnnotation type

lighthouse_image = testimage("lighthouse")
lighthouse_image_scaled = imresize(lighthouse_image, (640, 640))

annotations = annotate(lighthouse_image_scaled, annotator)
# 1-element Vector{BoundingBoxAnnotation{Int64, Float32}}:
#  BoundingBoxAnnotation{Int64, Float32}(GeometryBasics.HyperRectangle{2, Float32}(Float32[229.6118, 359.71802], Float32[8.796692, 21.370972]), ImageAnnotation{Int64}(0, 0.46368203f0, "ONNXRuntimeObjectAnnotator"))