Digital image fill

1. In a digital medium environment to fill a hole in a digital image, a method implemented by a computing device, the method comprising:

generating, by the computing device, an attention map based on the digital image and a guidance digital image, the attention map describing which part of the guidance digital image is to be used as a basis to fill the hole in the digital image;

filtering, by the computing device, a second feature map, generated from the guidance digital image using machine learning, based on the attention map;

combining, by the computing device, the filtered second feature map with a first feature map generated from the digital image using machine learning; and

generating, by the computing device, a filled digital image by decoding the filtered second feature map combined with the first feature map.

2. The method as described in claim 1 , further comprising generating the first feature map.

3. The method as described in claim 2 , wherein the generating of the first feature map is performed using a convolutional neural network and further comprising generating the second feature map using a convolutional neural network having a visual geometry group (VGG) architecture.

4. The method as described in claim 1 , wherein the generating of the attention map is performed using a convolutional neural network as part of machine learning.

5. The method as described in claim 1 , wherein the combining is performed using summation or concatenation.

6. The method as described in claim 1 , further comprising calculating an amount of error using a loss function by a generative adversarial network (GAN).

7. The method as described in claim 1 , further comprising calculating an amount of error based on a discriminator implemented by a neural network or perceptual loss.

8. The method as described in claim 1 , wherein the generating of the filled digital image using machine learning is performed by a deconvolution neural network by decoding the combined filtered second feature map, the first feature map, and at least a portion of the second feature map obtained via a skip layer from a convolutional neural network used to generate the second feature map using machine learning.

9. In a digital medium environment to fill a hole in a digital image, a computing device comprising:

a processing system; and

a computer-readable storage medium having instructions stored thereon that, responsive to execution by the processing system, causes the processing system to perform operations comprising:

generating an attention map based on the digital image and a guidance digital image, the attention map describing which a part of the guidance digital image to be used as a basis to fill the hole in the digital image;

filtering a second feature map based on the attention map, the second feature map generated using machine learning from the guidance digital image;

combining the filtered second feature map with a first feature map, the first feature map generating using machine learning from the digital image having to hole to be filled; and

generating a filled digital image by decoding the combined filtered second feature map and first feature map.

10. The computing device as described in claim 9 , the operations further comprising generating a search query to locate the guidance digital image based on a search of a digital image repository using features learned from the digital image having the hole to be filed.

11. The computing device as described in claim 9 , wherein the first feature map is generated using a convolutional neural network and the second feature map is generated using a convolutional neural network having a visual geometry group (VGG) architecture.

12. The computing device as described in claim 9 , wherein the generating of the attention map is performed using a convolutional neural network as part of machine learning.

13. In a digital medium environment to fill a hole in a digital image, a system comprising:

means for generating an attention map based on the digital image and a guidance digital image, the attention map describing which part of the guidance digital image is to be used as a basis to fill the hole in the digital image;

means for filtering a second feature map, generated from the guidance digital image using machine learning, based on the attention map;

means for combining the filtered second feature map with a first feature map generated from the digital image using machine learning; and

means for generating a filled digital image by decoding the filtered second feature map combined with the first feature map.

14. The system as described in claim 13 , wherein the attention map generating means generates the first feature map.

15. The system as described in claim 13 , further comprising means for generating the first feature map using a convolutional neural network and means for generating the second feature map using a convolutional neural network having a visual geometry group (VGG) architecture.

16. The system as described in claim 13 , wherein the attention map generating means includes a convolutional neural network.

17. The system as described in claim 13 , wherein the combining means includes means for summation or means for concatenation.

18. The system as described in claim 13 , further comprising means for calculating an amount of error using a loss function by a generative adversarial network (GAN).

19. The system as described in claim 13 , further comprising means for calculating an amount of error based on a discriminator implemented by a neural network or perceptual loss.

20. The system as described in claim 13 , wherein the filled digital image generating means includes a deconvolution neural network to decode the combined filtered second feature map, the first feature map, and at least a portion of the second feature map obtained via a skip layer from a convolutional neural network used to generate the second feature map using machine learning.