Split machine-learning assisted graphics rendering pipeline

1. A method comprising:

rendering, on a graphics processing unit (GPU), a low-resolution image associated with a scene, the low-resolution image having a resolution that is lower than a target resolution;

transmitting a version of the low-resolution image to a neural accelerator;

processing, on the neural accelerator, the version of the low-resolution image using a trained machine-learning model, thereby outputting a plurality of control parameters;

transmitting the control parameters from the neural accelerator to the GPU;

processing, on the GPU, the low-resolution image and the control parameters to construct a high-resolution image having the target resolution, wherein the GPU is programmed to determine a plurality of pixel weights for performing an interpolation using the control parameters; and

outputting the high-resolution image.

2. The method of claim 1 , wherein:

each of the low-resolution image and the high-resolution image comprise a plurality of pixels, respectively; and

the control parameters are used for determining, for each pixel of the high-resolution image, the one or more pixel weights respectively associated with one or more specific pixels of the low-resolution image that are used for computing a value for that pixel of the high-resolution image.

3. The method of claim 1 , wherein the pixel weights collectively indicate a respective orientation of one or more edges detected by the trained machine learning neural model in the version of the low-resolution image.

4. The method of claim 1 , wherein each of the low-resolution image and the high-resolution image comprise a color image, respectively, the method further comprising:

processing, on the GPU, the low-resolution image to generate the version of the low-resolution image;

wherein the version of the low-resolution image comprises a monochrome (one-channel) image.

5. The method of claim 1 , further comprising:

processing, on the GPU, the high-resolution image and additional data associated with the scene, thereby outputting a composite image; and

displaying the composite image on a device display.

6. The method of claim 5 , further comprising:

transmitting the composite image to a display processing unit (DPU); and

executing, on the DPU, display correction and scaling of the composite image before displaying the composite image on the device display.

7. The method of claim 1 , wherein the trained machine-learning model is programmed as a convolutional neural network (CNN).

8. A system comprising: one or more processors; and a non-transitory memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to perform a method, the method comprising:

rendering, on a graphics processing unit (GPU), a low-resolution image associated with a scene, the low-resolution image having a resolution that is lower than a target resolution;

transmitting a version of the low-resolution image to a neural accelerator;

processing, on the neural accelerator, the version of the low-resolution image using a trained machine-learning model, thereby outputting a plurality of control parameters;

transmitting the control parameters from the neural accelerator to the GPU;

processing, on the GPU, the low-resolution image and the control parameters to construct a high-resolution image having the target resolution, wherein the GPU is programmed to determine a plurality of pixel weights for performing an interpolation using the control parameters; and

outputting the high-resolution image.

9. The system of claim 8 , wherein:

each of the low-resolution image and the high-resolution image comprise a plurality of pixels, respectively; and

the control parameters are used for determining, for each pixel of the high-resolution image, the one or more pixel weights respectively associated with one or more specific pixels of the low-resolution image that are used for computing a value for that pixel of the high-resolution image.

10. The system of claim 8 , wherein the pixel weights collectively indicate a respective orientation of one or more edges detected by the trained machine learning neural model in the version of the low-resolution image.

11. The system of claim 8 , wherein each of the low-resolution image and the high-resolution image comprise a color image, respectively, and wherein the processors are further operable when executing the instructions to perform:

processing, on the GPU, the low-resolution image to generate the version of the low-resolution image;

wherein the version of the low-resolution image comprises a monochrome (one-channel) image.

12. The system of claim 8 , wherein the processors are further operable when executing the instructions to perform:

processing, on the GPU, the high-resolution image and additional data associated with the scene, thereby outputting a composite image; and

displaying the composite image on a device display.

13. The system of claim 12 , wherein the processors are further operable when executing the instructions to perform:

transmitting the composite image to a display processing unit (DPU); and

executing, on the DPU, display correction and scaling of the composite image before displaying the composite image on the device display.

14. The system of claim 8 , wherein the trained machine-learning model is programmed as a convolutional neural network (CNN).

15. One or more computer-readable non-transitory storage media embodying software that is operable when executed to perform a method, the method comprising:

rendering, on a graphics processing unit (GPU), a low-resolution image associated with a scene, the low-resolution image having a resolution that is lower than a target resolution;

transmitting a version of the low-resolution image to a neural accelerator;

processing, on the neural accelerator, the version of the low-resolution image using a trained machine-learning model, thereby outputting a plurality of control parameters;

transmitting the control parameters from the neural accelerator to the GPU;

processing, on the GPU, the low-resolution image and the control parameters to construct a high-resolution image having the target resolution, wherein the GPU is programmed to determine a plurality of pixel weights for performing an interpolation using the control parameters; and

outputting the high-resolution image.

16. The storage media of claim 15 , wherein:

each of the low-resolution image and the high-resolution image comprise a plurality of pixels, respectively; and

the control parameters are used for determining, for each pixel of the high-resolution image, the one or more pixel weights respectively associated with one or more specific pixels of the low-resolution image that are used for computing a value for that pixel of the high-resolution image.

17. The storage media of claim 15 , wherein the pixel weights collectively indicate a respective orientation of one or more edges detected by the trained machine learning neural model in the version of the low-resolution image.

18. The storage media of claim 15 , wherein each of the low-resolution image and the high-resolution image comprise a color image, respectively, and wherein the software is further operable when executed to perform:

processing, on the GPU, the low-resolution image to generate the version of the low-resolution image;

wherein the version of the low-resolution image comprises a monochrome (one-channel) image.

19. The storage media of claim 15 , wherein the software is further operable when executed to perform:

processing, on the GPU, the high-resolution image and additional data associated with the scene, thereby outputting a composite image; and

displaying the composite image on a device display.

20. The storage media of claim 19 , wherein the software is further operable when executed to perform:

transmitting the composite image to a display processing unit (DPU); and

executing, on the DPU, display correction and scaling of the composite image before displaying the composite image on the device display.