Object data curation of map information using neural networks for autonomous systems and applications

A system comprising:

• one or more sensors;
• and
• one or more processors to:
• determine, based at least on map data, one or more object locations associated with a target object;
• determine, using the object locations and location data associated with representations of sensor data, a subset of the representations of the sensor data corresponding to the one or more object locations;
• compute, for at least one sensor data representation of the subset of the representations of the sensor data, a feature vector corresponding to an instance of the target object depicted in the at least one sensor data representation;
• compute a similarity score between the feature vector and a template feature vector corresponding to a template of the target object;
• based at least on the similarity score being greater than a threshold similarity score, generate training data based at least on the at least one sensor data representation;
• and
• update one or more weights of a machine learning model using the training data.

A method comprising:

• determining, using map data, one or more object locations associated with a target object of interest;
• mapping the one or more object locations from the map data to one or more frames of sensor data captured at one or more vehicle locations corresponding to the one or more object locations;
• computing a similarity score for at least one frame of the one or more frames of the sensor data based at least on comparing a feature vector associated with an instance of the target object of interest depicted in the at least one frame to a template feature vector associated with a template of the target object of interest;
• ranking the at least one frame of the one or more frames of the sensor data based at least on the similarity score;
• assigning, based at least on the ranking of the at least one frame of the one or more frames of the sensor data, a label to the at least one frame of the one or more frames of the sensor data, the label indicating a classification associated with the target object of interest;
• and
• updating one or more weights of a machine learning model using the label as ground truth data and the at least one frame.