Training deep neural networks with synthetic images

1. A system comprising a computer including a processor and a memory, the memory including instructions such that the processor is programmed to:

generate, via a deep neural network, a first synthetic image based on a machine-generated simulated image;

generate a segmentation mask based on the synthetic image;

compare the segmentation mask with a ground truth mask of the synthetic image;

update the deep neural network based on the comparison; and

generate, via the updated deep neural network, a second synthetic image based on the simulated image.

2. The system of claim 1 , wherein the processor is further programmed to:

compare, via a Siamese neural network, the segmentation mask with the ground truth mask.

3. The system of claim 1 , wherein the simulated image depicts a plurality of objects.

4. The system of claim 3 , wherein the segmentation mask comprises a binary mask that classifies similar objects of the plurality objects as a single instance.

5. The system of claim 1 , wherein the simulated image is generated by a gaming engine.

6. The system of claim 1 , wherein the deep neural network comprises a deconvolutional neural network.

7. The system of claim 1 , wherein the processor is further programmed to:

determine a contrastive loss based on the comparison; and

update the deep neural network based on the contrastive loss.

8. The system of claim 1 , wherein the processor is further programmed to:

update at least one weight associated with a neuron of the deep neural network.

9. The system of claim 1 , wherein the processor is further programmed to:

evaluate the second synthetic image; and

generate a prediction indicative of whether the second synthetic image is machine-generated or is sourced from a real data distribution.

10. The system of claim 9 , wherein the processor is further programmed to:

generate feedback based on the prediction; and

update at least one weight associated with a neuron of the deep neural network when the prediction indicates that the second synthetic image is machine-generated.

11. The system of claim 9 , wherein the processor is further programmed to:

evaluate the second synthetic image via a convolutional neural network.

12. A method comprising:

generating, via a deep neural network, a first synthetic image based on a machine-generated simulated image;

generating a segmentation mask based on the synthetic image;

comparing the segmentation mask with a ground truth mask of the synthetic image;

updating the deep neural network based on the comparison; and

generating, via the updated deep neural network, a second synthetic image based on the simulated image.

13. The method of claim 12 , further comprising:

comparing, via a Siamese neural network, the segmentation mask with the ground truth mask.

14. The method of claim 12 , wherein the simulated image depicts a plurality of objects.

15. The method of claim 14 , wherein the segmentation mask comprises a binary mask that classifies similar objects of the plurality of objects as a single instance.

16. The method of claim 12 , wherein the simulated image is generated by a gaming engine.

17. The method of claim 12 , wherein the deep neural network comprises a deconvolutional neural network.

18. The method of claim 12 , further comprising:

determining a contrastive loss based on the comparison; and

updating the deep neural network based on the contrastive loss.

19. The method of claim 12 , further comprising:

updating at least one weight associated with a neuron of the deep neural network.

20. The method of claim 12 , further comprising:

evaluating the second synthetic image; and

generating a prediction indicative of whether the second synthetic image is machine-generated or is sourced from a real data distribution.