Training using rendered images

1. A method, comprising:

for a set of iterations, randomly positioning a three-dimensional (3D) object model in a virtual space having randomly selected textures;

for the set of iterations, rendering a two-dimensional (2D) image of the 3D object model in the virtual space and a corresponding annotation image; and

training a machine learning model using the rendered 2D images and corresponding annotation images.

2. The method of claim 1 , further comprising, for the set of iterations, randomly positioning a viewpoint of the virtual space.

3. The method of claim 2 , wherein randomly positioning the viewpoint comprises positioning a virtual camera in the virtual space at a random position that is pointing at the 3D object model.

4. The method of claim 3 , further comprising validating that the virtual camera is within a boundary before rendering the 2D image.

5. The method of claim 1 , wherein randomly positioning the 3D object model comprises positioning the 3D object model above a surface with a random orientation.

6. The method of claim 5 , wherein randomly positioning the 3D object model comprises dropping the 3D object model onto the surface.

7. The method of claim 6 , further comprising determining that a position of the 3D object model has stabilized before rendering the 2D image of the 3D object model.

8. The method of claim 1 , further comprising, for the set of iterations, randomizing an illumination of the virtual space.

9. The method of claim 1 , further comprising, for the set of iterations, selecting a surface.

10. The method of claim 1 , further comprising providing, by a web service, the trained machine learning model.

11. An apparatus, comprising:

a memory; and

a processor coupled to the memory, wherein the processor is to:

generate a virtual space;

randomly select a texture for a surface in the virtual space;

randomly position a three-dimensional (3D) object model on the surface;

determine a position of a camera in the virtual space with a field of view that includes the 3D object model;

render a two-dimensional (2D) image from the position of the camera;

render an annotation image that indicates a location of the 3D object model in the 2D image; and

train a machine learning model using the 2D image and the annotation image.

12. The apparatus of claim 11 , wherein the processor is to produce a user interface including a control to select the 3D object model, to establish a set of textures, or to select the virtual space.

13. The apparatus of claim 11 , wherein the processor is to randomly position the 3D object model on the surface by calculating a fall of the 3D object model onto the surface from a random orientation above the surface.

14. A non-transitory tangible computer-readable medium storing executable code, comprising:

code to cause a processor to drop a virtual three-dimensional (3D) object from a random orientation onto a surface with a randomly selected texture in a virtual environment for a set of repetitions;

code to cause the processor to generate a two-dimensional (2D) training image and a 2D annotation image from a random viewpoint for each of the set of repetitions; and

code to cause the processor to provide a machine learning model that is trained based on the 2D training images and the 2D annotation images.

15. The computer-readable medium of claim 14 , further comprising code to cause the processor to randomize lighting in the virtual environment.