System and method using augmented reality for efficient collection of training data for machine learning

1. A computer system for facilitating efficient collection of training data, the system comprising:

a processor; and

a storage device storing instructions that when executed by the processor cause the processor to perform a method, the method comprising:

obtaining, by a recording device, a first image of a physical object in a scene which is associated with a three-dimensional (3D) world coordinate frame;

marking, by a user of the recording device on the first image, a plurality of vertices associated with the physical object, wherein a vertex has 3D coordinates based on the 3D world coordinate frame;

obtaining, by the recording device, a plurality of second images of the physical object in the scene while changing one or more characteristics of the scene;

displaying, by the recording device on a display of the recording device, a respective second image of the physical object in the scene by:

projecting, on the display, the marked plurality of vertices associated with the physical object onto the respective second image; and

indicating, on the display, a two-dimensional (2D) bounding area which is associated with the physical object and includes the marked vertices;

storing, in a collection of training data, the first image with the marked plurality of vertices and the plurality of second images with the projected marked vertices; and

training, based on the collection of training data, a convolutional neural network to identify features of the physical object.

2. The computer system of claim 1 , wherein the marked plurality of vertices corresponds to one or more regions of interest on the physical object, and

wherein projecting the marked vertices further comprising indicating a 2D bounding area associated with the one or more regions of interest on the physical object.

3. The computer system of claim 1 , wherein the marked plurality of vertices can indicate one or more of:

a polygon;

a portion of a surface plane; and

a volume.

4. The computer system of claim 1 , wherein marking the plurality of vertices further comprises:

determining how to indicate the 2D bounding area of the projected marked vertices on the respective second image.

5. The computer system of claim 1 , wherein the 2D bounding area and the respective second image are presented on a display associated with the recording device, and

wherein the 2D bounding area indicates a 2D shape or a 3D volume.

6. The computer system of claim 1 , wherein the 2D bounding area is indicated by one or more of:

a type, pattern, or color of a connector between the projected vertices in the respective second image;

a color, shading, or fill of a shape formed by connecting the projected vertices in the respective second image;

text describing the 2D bounding area; and

an indication of a label or description for the 2D bounding area.

7. The computer system of claim 1 , wherein a 2D bounding area corresponds to a characteristic of the scene.

8. The computer system of claim 1 , wherein a characteristic of the scene is one or more of:

a pose of the recording device;

a lighting of the scene;

a distance of the recording device from the physical object;

an orientation of the recording device in relation to the physical object;

a background of the physical object or the scene; and

an occlusion of one or more portions of the physical object.

9. The computer system of claim 1 , wherein the method further comprises:

deploying the trained network on a mobile computing device to identify the features of the physical object.

10. The computer system of claim 1 , wherein the recording device includes one or more of:

an augmented reality device;

a virtual reality device;

a device with magnetic sensors which determine 3D coordinates for a vertex in the 3D world coordinate frame;

a camera and a hand-tracking sensor;

a camera which records red, green, and blue (RGB), wherein the hand-tracking sensor determines 3D coordinates for a vertex in the 3D world coordinate frame;

a camera which records red, green, and blue (RGB), and a 3D sensor which records a depth;

a device which records images or video, and determines 3D coordinates for a vertex in the 3D world coordinate frame based on visual cues or a position-sensing technology; and

a device which records images or video and includes a (3D) sensor.