System and method for capture and adaptive data generation for training for machine vision

A computer-implemented method of performing machine vision prediction of digital images using synthetically generated training assets, the method implemented using a first set of executable computer program instructions, the method comprising:

• digitally capturing a plurality of assets, each of the assets comprising a plurality of digital data files;
• parameterizing each of the assets, to create parameterized assets, by an asset state comprising a first scale, a first plurality of material properties, a first deformation, a first mesh parameterization, and first camera points of interests;
• parameterizing each scene among the assets by a first list of the assets and a second list of cameras for scene attributes of a second scale, a second plurality of material properties, a second deformation, a second mesh parameterization, and second camera points of interests;
• selecting a dataset target from among training, validating and testing, resulting in a selected dataset target;
• tagging each scene for the selected dataset target;
• configuring an environment map for the selected dataset target;
• configuring annotation output states to provide annotations for different outputs;
• configuring output images comprising one or more masks selected from among object identifiers and segmentation, depth, color, or custom masks;
• configuring output metadata comprising, positions and alignment, labels, hierarchy;
• configuring quality settings for output comprising render samples per pixel and resolution;
• each of the foregoing comprising minimum and maximum values and collectively comprising a parameter space of the assets;
• using a plurality of different parameterized assets from among all the parameterized assets, configuring a plurality of machine learning trials and executing the trials to train a machine vision model, resulting in creating and storing a trained machine vision model;
• scoring the trained machine vision model to compute performance of the different parameterized assets, resulting in a composite score value for comparison against a set of success criteria and a set of points of failure;
• clustering one or more new data points around the points of failure to resample the parametric space using any one of:
• single choice sampling, sequential choice sampling, local cluster sampling, and local spatial sampling[MF4] with a probability distribution function;
• and
• repeating the configuring the plurality of machine learning trials, executing the trials, scoring the trained machine vision model and the clustering until the composite score value satisfies the success criteria.

One or more non-transitory computer-readable storage media storing instructions programmed for performing machine vision prediction of digital images using synthetically generated training assets and which instructions, when executed using one or more processors, cause the one or more processors to execute:

• digitally capturing a plurality of assets, each of the assets comprising a one or more digital data files;
• configuring each of the assets in the plurality of assets with a plurality of asset attributes;
• under computer program control, selecting a plurality of different combinations of parameters from among the plurality of asset attributes, and creating a plurality of sets of different synthetic dataset parameters;
• using computer graphics software that is different than the first set of executable computer program instructions, and example parameter values from among the synthetic dataset parameters, creating a synthetic dataset example image by compiling from a plurality of example images and metadata;
• configuring a plurality of machine learning trials and executing the trials to train a machine vision model, resulting in creating and storing a trained machine vision model;
• executing a validation of the trained machine vision model;
• inferring a prediction using the trained machine vision model.