Three-dimensional detection and tracking pipline recommendation using performance prediction

A method comprising:

• storing, in a first memory area, at least two regressors, each regressor being trained to predict performances of (1) a first pipeline of an object detection (OD) and pose estimation (PE) algorithm and (2) a second pipeline of an OD and PE algorithm different from the OD and PE algorithm of the first pipeline, in response to a first feature vector corresponding to a first object as input to the regressor, the first and second pipelines being different from each other;
• storing, in a second memory area, a second feature vector corresponding to a second object;
• running the regressors with the second feature vector as input common to the regressors to predict the performances of the pipelines of the different OD and PE algorithms;
• indicating, on a display, a recommendation of at least one pipeline of the OD and PE algorithms comprising the corresponding first pipeline and the corresponding second pipeline based on the predicted performances;
• and
• deriving a pose of the second object by using the recommended pipeline.

A non-transitory computer readable medium storing instructions, when executed by one or more processors, to perform a method comprising:

• storing, in a first memory area, training data corresponding to an object and predetermined flag data indicating a pipeline of an object detection (OD) and pose estimation (PE) algorithm for use with the training data;
• receiving image frames including the object in a real scene captured by a camera;
• selecting one pipeline from among a plurality of pipelines of different OD and PE algorithms stored in a second memory area based on the flag data without using the image frames including the object;
• and
• deriving, with the selected pipeline, a pose of the object with respect to the camera, using the image frames and the training data.