Short-term and long-term memory on an edge device

A method comprising:

• receiving visual data at a device, wherein the visual data is captured at a first time at a camera that is proximate to the device, wherein the camera is affixed to a first vehicle, and wherein a traffic device or landmark is visible in the visual data;
• storing the visual data at a first resolution in a memory of the device;
• processing the visual data at a processor coupled to the memory to produce an observations data;
• determining an uncertainty of a position of the traffic device or landmark in a map;
• determining a priority of the visual data, wherein the priority of the visual data is based at least in part on the uncertainty and on the observations data;
• and
• storing the visual data at a second resolution in the memory based at least in part on the priority.

A method comprising:

• receiving visual data at a device, wherein the visual data is captured at a first time at a camera that is proximate to the device, wherein the camera is affixed to a first vehicle;
• wherein a second vehicle is visible in the visual data;
• storing the visual data at a first resolution in a memory of the device;
• processing the visual data at a processor coupled to the memory to produce an observations data;
• wherein the observations data comprises a descriptor of the second vehicle;
• wherein the descriptor of the second vehicle comprises an indication that the second vehicle may have been involved in a rare traffic event;
• determining a priority of the visual data based at least in part on the observations data;
• and
• storing the visual data at a second resolution in the memory based at least in part on the priority;
• determining a second priority based on the observations data and based at least in part on the indication;
• and
• deleting the stored visual data at the second resolution from the memory based on the second priority.

A method comprising:

• receiving visual data at a device, wherein the visual data is captured at a first time at a camera that is proximate to the device;
• storing the visual data at a first resolution in a memory of the device;
• determining a priority of the visual data;
• storing the visual data at a second resolution in the memory based at least in part on the priority, wherein a first portion of the memory contains visual data stored at the first resolution, and wherein a second portion of the memory contains visual data stored at the second resolution;
• receiving an inference model at the device;
• receiving an indication of an input resolution associated with the inference model;
• selecting a stored visual data from at least one of the first portion of the memory and the second portion of the memory based at least in part on the indicated input resolution;
• and
• processing the selected visual data with the inference model at a processor coupled to the memory to produce an inference data.