Three dimensional bounding box estimation from two dimensional images

A method comprising:

• inputting sensor data associated with an environment including an object into a machine learning algorithm that comprises:
• a coarse output branch configured to output a coarse output;
• and
• a fine offset branch configured to output a numeric offset with respect to the coarse output by the coarse output branch;
• receiving, from the machine learning algorithm, output defining an orientation of a three dimensional bounding box associated with the object, wherein the output comprises:
• the coarse output, wherein the coarse output represents a coarse orientation of the three dimensional bounding box;
• the numeric offset, wherein the numeric offset represents an orientation offset with respect to the coarse orientation of the three dimensional bounding box;
• a confidence value of a set of confidence values associated with the coarse output;
• and
• a dimension of the three dimensional bounding box.

One or more non-transitory computer readable media comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

• inputting sensor data associated with an environment including an object characterized by a first plurality of kinematic variables into a machine learning algorithm, wherein the machine learning algorithm comprises:
• a coarse output branch configured to output a coarse output, wherein the coarse output estimates a value of a physical property associated with the object;
• and
• a fine offset branch configured to output a numeric offset with respect to the coarse output by the coarse output branch;
• and
• receiving, from the machine learning algorithm, reparameterization output associated with the object traversing the environment, the reparameterization output reparameterizing the first plurality of kinematic variables into a second plurality of kinematic variables,
• wherein the second plurality of kinematic variables comprises fewer kinematic variables than the first plurality of kinematic variables, and
• further wherein the reparameterization output comprises the coarse output, the numeric offset, and a confidence value of a set of confidence values associated with the coarse output.