System and method for generating multi-exposure frames from single input

1. A method comprising:

obtaining, using at least one image sensor of an electronic device, a first image frame of a scene;

using a convolutional neural network to generate, from the first image frame, multiple second image frames simulated to have different exposures, wherein one or more objects in the scene in each second image frame are aligned with one or more corresponding objects in the scene in at least one other second image frame and are aligned with one or more corresponding objects in the scene in the first image frame; and

blending the multiple second image frames to generate a final image of the scene.

2. The method of claim 1 , wherein the multiple second image frames comprise:

a first generated image frame simulated to have a specified exposure;

a second generated image frame simulated to have a second exposure that is longer than the specified exposure; and

a third generated image frame simulated to have a third exposure that is shorter than the specified exposure.

3. The method of claim 2 , wherein the convolutional neural network performs operations in a common encoder path for all of the multiple second image frames and performs distinct operations in different decoder paths for different ones of the multiple second image frames.

4. The method of claim 3 , wherein each of the decoder paths comprises at least one decoder stage that includes an upsampling layer, a convolutional layer, and a separable convolutional layer.

5. The method of claim 3 , wherein:

the decoder path for the second generated image frame comprises more decoder stages than the decoder path for the first generated image frame; and

the decoder path for the third generated image frame comprises fewer decoder stages than the decoder path for the first generated image frame.

6. The method of claim 1 , wherein:

the first image frame of the scene is in a raw format;

the multiple second image frames are in a YUV format; and

the convolutional neural network performs raw-to-YUV conversion.

7. The method of claim 1 , wherein blending the multiple second image frames to generate the final image comprises:

performing at least one of a motion blur reduction (MBR) deghosting operation, a MBR blending operation, or a high dynamic range (HDR) blending operation using at least some of the multiple second image frames.

8. The method of claim 1 , wherein the convolutional neural network is trained by:

obtaining multiple initial image frames of a scene including multiple target image frames;

generating output images;

computing a loss function indicating differences between the target image frames and the generated output images;

updating weights associated with parameters of the convolutional neural network based on the loss function; and

repeating the generating, computing, and updating operations one or more times in an iterative manner.

9. An electronic device comprising:

at least one image sensor; and

at least one processing device configured to:

obtain a first image frame of a scene using the at least one image sensor;

use a convolutional neural network to generate, from the first image frame, multiple second image frames simulated to have different exposures, wherein one or more objects in the scene in each second image frame are aligned with one or more corresponding objects in the scene in at least one other second image frame and are aligned with one or more corresponding objects in the scene in the first image frame; and

blend the multiple second image frames to generate a final image of the scene.

10. The electronic device of claim 9 , wherein the multiple second image frames comprise:

a first generated image frame simulated to have a specified exposure;

a second generated image frame simulated to have a second exposure that is longer than the specified exposure; and

a third generated image frame simulated to have a third exposure that is shorter than the specified exposure.

11. The electronic device of claim 10 , wherein the convolutional neural network is configured to perform operations in a common encoder path for all of the multiple second image frames and to perform distinct operations in different decoder paths for different ones of the multiple second image frames.

12. The electronic device of claim 11 , wherein each of the decoder paths comprises at least one decoder stage that includes an upsampling layer, a convolutional layer, and a separable convolutional layer.

13. The electronic device of claim 11 , wherein:

the decoder path for the second generated image frame comprises more decoder stages than the decoder path for the first generated image frame; and

the decoder path for the third generated image frame comprises fewer decoder stages than the decoder path for the first generated image frame.

14. The electronic device of claim 9 , wherein:

the first image frame of the scene is in a raw format;

the multiple second image frames are in a YUV format; and

the convolutional neural network is configured to perform raw-to-YUV conversion.

15. The electronic device of claim 9 , wherein, to blend the multiple second image frames to generate the final image, the at least one processing device is configured to perform at least one of a motion blur reduction (MBR) deghosting operation, a MBR blending operation, or a high dynamic range (HDR) blending operation using at least some of the multiple second image frames.

16. The electronic device of claim 9 , wherein the convolutional neural network is trained by:

obtaining multiple initial image frames of a scene including multiple target image frames;

generating output images similar to the target image frames;

computing a loss function indicating differences between the target image frames and the generated output images;

updating weights associated with parameters of the convolutional neural network based on the loss function; and

repeating the generating, computing, and updating operations one or more times in an iterative manner.

17. A non-transitory machine-readable medium containing instructions that when executed cause at least one processor of an electronic device to:

obtain a first image frame of a scene using at least one image sensor of the electronic device;

use a convolutional neural network to generate, from the first image frame, multiple second image frames simulated to have different exposures, wherein one or more objects in the scene in each second image frame are aligned with one or more corresponding objects in the scene in at least one other second image frame and are aligned with one or more corresponding objects in the scene in the first image frame; and

blend the multiple second image frames to generate a final image of the scene.

18. The non-transitory machine-readable medium of claim 17 , wherein the multiple second image frames comprise:

a first generated image frame simulated to have a specified exposure;

a second generated image frame simulated to have a second exposure that is longer than the specified exposure; and

a third generated image frame simulated to have a third exposure that is shorter than the specified exposure.

19. The non-transitory machine-readable medium of claim 18 , wherein the convolutional neural network is configured to perform operations in a common encoder path for all of the multiple second image frames and to perform distinct operations in different decoder paths for different ones of the multiple second image frames.

20. The non-transitory machine-readable medium of claim 19 , wherein each of the decoder paths comprises at least one decoder stage that includes an upsampling layer, a convolutional layer, and a separable convolutional layer.