Systems and methods for detecting objects within an image in a wide-view format

1. A detection system for locating an object, comprising:

a processor; and

a memory storing instructions that, when executed by the processor, cause the processor to:

convert a real image in a wide-view format to a synthetic representation using a style model, wherein the synthetic representation depicts a distorted view of the object;

identify features of the object using an extraction model that distinguishes different scales of the synthetic representation and a simulated scene to define structures associated with the distorted view;

detect the object using a decoder model that identifies an attribute and a bounding box of the object from the features; and

execute a task using the attribute and the bounding box to localize the object in the simulated scene.

2. The detection system of claim 1 , further including instructions to:

output by the decoder model a synthetic location of the object in the simulated scene, wherein the synthetic location is represented by pixels; and

compare the synthetic location to geometries of the real image using a transformation matrix to derive an actual location and generate a map for visualizing the actual location.

3. The detection system of claim 2 , further including instructions to generate the map using the attribute, the bounding box, and feature predictions from a prior location of the object.

4. The detection system of claim 2 , further including instructions to derive the transformation matrix according to the real image being taken by a camera in a fixed position.

5. The detection system of claim 1 , further including instructions to train the style model using a simulated environment having simulated images from a camera in a fixed position as ground truths, wherein the simulated images are in the wide-view format and distorted.

6. The detection system of claim 1 , further including instructions to:

localize the object by a class that labels by type; and

generate a map having the object on a road that is simulated using the bounding box and the class.

7. The detection system of claim 1 , wherein the extraction model is a pyramid encoder and the decoder model is a neural network.

8. The detection system of claim 1 , wherein the task is one of navigating and automating a device.

9. A non-transitory computer-readable medium for locating an object, comprising:

instructions that when executed by a processor cause the processor to:

convert a real image in a wide-view format to a synthetic representation using a style model, wherein the synthetic representation depicts a distorted view of the object;

identify features of the object using an extraction model that distinguishes different scales of the synthetic representation and a simulated scene to define structures associated with the distorted view;

detect the object using a decoder model that identifies an attribute and a bounding box of the object from the features; and

execute a task using the attribute and the bounding box to localize the object in the simulated scene.

10. The non-transitory computer-readable medium of claim 9 , further including instructions to:

output by the decoder model a synthetic location of the object in the simulated scene, wherein the synthetic location is represented by pixels; and

compare the synthetic location to geometries of the real image using a transformation matrix to derive an actual location and generate a map for visualizing the actual location.

11. The non-transitory computer-readable medium of claim 10 , further including instructions to generate the map using the attribute, the bounding box, and feature predictions from a prior location of the object.

12. The non-transitory computer-readable medium of claim 10 , further including instructions to derive the transformation matrix according to the real image being taken by a camera in a fixed position.

13. A method comprising:

converting a real image in a wide-view format to a synthetic representation using a style model, wherein the synthetic representation depicts a distorted view of an object;

identifying features of the object using an extraction model that distinguishes different scales of the synthetic representation and a simulated scene to define structures associated with the distorted view;

detecting the object using a decoder model that identifies an attribute and a bounding box of the object from the features; and

executing a task using the attribute and the bounding box to localize the object in the simulated scene.

14. The method of claim 13 , further comprising:

outputting by the decoder model a synthetic location of the object in the simulated scene, wherein the synthetic location is represented by pixels; and

comparing the synthetic location to geometries of the real image using a transformation matrix to derive an actual location and generating a map for visualizing the actual location.

15. The method of claim 14 , further comprising generating the map using the attribute, the bounding box, and feature predictions from a prior location of the object.

16. The method of claim 14 , further comprising deriving the transformation matrix according to the real image being taken by a camera in a fixed position.

17. The method of claim 13 , further comprising training the style model using a simulated environment having simulated images from a camera in a fixed position as ground truths, wherein the simulated images are in the wide-view format and distorted.

18. The method of claim 13 , further comprising:

localizing the object by a class that labels by type; and

generating a map having the object on a road that is simulated using the bounding box and the class.

19. The method of claim 13 , wherein the extraction model is a pyramid encoder and the decoder model is a neural network.

20. The method of claim 13 , wherein the task is one of navigating and automating a device.