Real time holography using learned error feedback

1. A heads up display (HUD) system comprising:

a holographic plane; and

a system for generating holographic images for presentation at the holographic plane, the system comprising circuitry configured to:

apply a machine learning model to a target holographic image to generate a feedback strength value;

apply an iterative propagation feedback model based on the target holographic image and the feedback strength value to generate a phase diffraction pattern; and

output the phase diffraction pattern for presentation at the holographic plane for generation of a corresponding holographic image.

2. The HUD system of claim 1 , wherein to apply the iterative propagation feedback model based on the target holographic image and the feedback strength value, the circuitry is configured to:

receive a current phase and amplitude image plane estimation; and

replace an amplitude plane of the current phase and amplitude image plane estimation with a next target amplitude image comprising:

a sum of the target holographic image and a product of the feedback strength value and a correction term comprising a difference between the target holographic image and the current phase and amplitude image plane to generate a current amplitude adjusted image.

3. The HUD system of claim 2 , wherein to apply the iterative propagation feedback model based on the target holographic image and the feedback strength value, the circuitry is configured to:

apply an inverse propagation model to a previous amplitude adjusted image to generate a phase and amplitude holographic plane estimation;

normalize an amplitude plane of the phase and amplitude holographic plane estimation to generate a phase only diffraction pattern image; and

apply a forward propagation model to the phase only diffraction pattern image to generate the current phase and amplitude image plane estimation.

4. The HUD system of claim 1 , wherein the machine learning model corresponds to a trained neural network and the machine learning model comprises at least one of fewer convolutional kernels with respect to the trained neural network or a reduced bit depth of weights of the machine learning model with respect to the trained neural network.

5. The HUD system of claim 4 , wherein the target holographic image is downsampled prior to application of the machine learning model.

6. The HUD system of claim 1 , wherein the target holographic image comprises first and second color planes, and the circuitry is configured to apply the machine learning model to the target holographic image to generate the feedback strength value for the first color plane and a second feedback strength value for the second color plane.

7. The HUD system of claim 6 , wherein the circuitry is further to:

apply the iterative propagation feedback model based on the second color plane of the target holographic image; and

use the second feedback strength value to generate a second phase diffraction pattern for presentation at the holographic plane or a second holographic plane.

8. The HUD system of claim 1 , wherein the holographic image is a three dimensional holographic image.

9. A method for generating holographic images for presentation at a holographic plane of a heads up display, the method comprising:

applying, via one or more processors, a machine learning model to a target holographic image to generate a feedback strength value;

applying an iterative propagation feedback model based on the target holographic image and the feedback strength value to generate a phase diffraction pattern; and

outputting the phase diffraction pattern for presentation at the holographic plane for generation of a corresponding holographic image.

10. The method of claim 9 , wherein applying the iterative propagation feedback model based on the target holographic image and the feedback strength value comprises:

receiving a current phase and amplitude image plane estimation;

replacing an amplitude plane of the current phase and amplitude image plane estimation with a next target amplitude image comprising a sum of the target holographic image and a product of the feedback strength value and a correction term comprising a difference between the target holographic image and the current phase and amplitude image plane to generate a current amplitude adjusted image;

applying an inverse propagation model to a previous amplitude adjusted image to generate a phase and amplitude holographic plane estimation;

normalizing an amplitude plane of the phase and amplitude holographic plane estimation to generate a phase only diffraction pattern image; and

applying a forward propagation model to the phase only diffraction pattern image to generate the current phase and amplitude image plane estimation.

11. The method of claim 9 , wherein the machine learning model corresponds to a trained neural network and the machine learning model comprises at least one of fewer convolutional kernels with respect to the trained neural network or a reduced bit depth of weights of the machine learning model with respect to the trained neural network.

12. The method of claim 11 , further comprising downsampling the target holographic image prior to applying the machine learning model.

13. The method of claim 9 , wherein the target holographic image comprises first and second color planes and the method further comprises applying the machine learning model to the target holographic image to generate the feedback strength value for the first color plane and a second feedback strength value for the second color plane.

14. The method of claim 13 , further comprising:

applying the iterative propagation feedback model based on the second color plane of the target holographic image; and

using the second feedback strength value to generate a second phase diffraction pattern for presentation at the holographic plane or a second holographic plane.

15. The method of claim 9 , wherein the holographic image is a three dimensional holographic image.

16. A non-transitory machine readable medium comprising a plurality of instructions that, in response to being executed by a processor of a holographic heads up display system, causes the processor to perform operations comprising:

applying a machine learning model to a target holographic image to generate a feedback strength value;

applying an iterative propagation feedback model based on the target holographic image and the feedback strength value to generate a phase diffraction pattern for presentation at a holographic plane of the holographic heads up display system; and

outputting the phase diffraction pattern for presentation at the holographic plane for generation of a corresponding holographic image.

17. The non-transitory machine readable medium as in claim 16 , wherein applying the iterative propagation feedback model based on the target holographic image and the feedback strength value includes operations comprising:

receiving a current phase and amplitude image plane estimation;

replacing an amplitude plane of the current phase and amplitude image plane estimation with a next target amplitude image comprising a sum of the target holographic image and a product of the feedback strength value and a correction term comprising a difference between the target holographic image and the current phase and amplitude image plane to generate a current amplitude adjusted image;

applying an inverse propagation model to a previous amplitude adjusted image to generate a phase and amplitude holographic plane estimation;

normalizing an amplitude plane of the phase and amplitude holographic plane estimation to generate a phase only diffraction pattern image; and

applying a forward propagation model to the phase only diffraction pattern image to generate the current phase and amplitude image plane estimation.

18. The non-transitory machine readable medium as in claim 16 , wherein the machine learning model corresponds to a trained neural network, the machine learning model comprises at least one of fewer convolutional kernels with respect to the trained neural network or a reduced bit depth of weights of the machine learning model with respect to the trained neural network, and the operations further comprise downsampling the target holographic image prior to applying the machine learning model.

19. The non-transitory machine readable medium as in claim 16 , wherein the target holographic image comprises first and second color planes and the operations further comprise applying the machine learning model to the target holographic image to generate the feedback strength value for the first color plane and a second feedback strength value for the second color plane.

20. The non-transitory machine readable medium as in claim 19 , wherein the holographic image is a three dimensional holographic image and the operations further comprise:

applying the iterative propagation feedback model based on the second color plane of the target holographic image; and

using the second feedback strength value to generate a second phase diffraction pattern for presentation at the holographic plane or a second holographic plane.