Inferrers

Photo by Laura Lee Moreau on Unsplash

Inferrers are modules that use either TensorFlow or OpenCV to find objects in images with a provided trained object detection model.

These inferrers are derived from the BaseInferrer class, which defines the methods that all inferrers need to have.

Documented below are these main methods.

TensorFlowInferrer(graph)

Initialize an inferrer with a path to the graph to load. The same applies to the TensorRTInferrer class.

Parameters:graph (str) – Relative path to the graph file.

OpenCVInferrer(graph, descriptor, input_size = 300)

OpenCV requires a descriptor file to interpret the graph properly. It also requires an image input size, but it has been set to a default of 300, since most models use this as their default image input size.

Parameters:

• graph (str) – Relative path to the graph file.
• descriptor (str) – Relative path to the descriptor file. See this link for more information.
• input_size (int) – Default 300. Set this to change the input blob size with OpenCV.

run(image)

Run inference on an image. All inferrers implement this method.

Parameters:image (np.ndarray) – Image to run inference on. Returns:Dictionary with two elements, num_detections, and detections, which is a list of dictionaries with the detection class, confidence and bounding box coordinates in the form of [top_x, top_y, bottom_x, bottom_y]. Return type:dict