Digital image fill
Fill techniques as implemented by a computing device are described to perform hole filling of a digital image. In one example, deeply learned features of a digital image using machine learning are used by a computing device as a basis to search a digital image repository to locate the guidance digital image. Once located, machine learning techniques are then used to align the guidance digital image with the hole to be filled in the digital image. Once aligned, the guidance digital image is then used to guide generation of fill for the hole in the digital image. Machine learning techniques are used to determine which parts of the guidance digital image are to be blended to fill the hole in the digital image and which parts of the hole are to receive new content that is synthesized by the computing device.
1. In a digital medium environment to fill a hole in a digital image using a guidance digital image, a system comprising:
a portion alignment module implemented at least partially in hardware of a computing device to generate parameters to align a portion of the guidance digital image with the hole to be filled in the digital image; and
a fill generation module implemented at least partially in hardware to transform the digital image by filling the hole in the digital image based at least in part on the parameters, the filling of the hole including:
blending the portion of the guidance digital image within a first part of the hole in the digital image; and
generating pixels using a model trained using machine learning for a second part of the hole in the digital image that is separate from the first part of the hole in the digital image.
2. The system as described in claim 1 , wherein the parameters specify at least one of a translation, a rotation, or a scaling to be applied to align the portion of the guidance digital image with respect to the hole to be filled in the digital image.
3. The system as described in claim 1 , wherein the generated pixels are based at least in part on pixels of the digital image with the hole to be filled.
4. The system as described in claim 3 , further comprising a ranking module to rank two or more said filled digital images based on an amount of realism exhibited by the two or more filled digital images.
5. The system as described in claim 1 , wherein the parameters are affine parameters.
6. The system as described in claim 1 , wherein the generated pixels are based at least in part on the portion of the guidance digital image.
7. The system as described in claim 1 , wherein the portion alignment module generates the parameters using a model trained using machine learning.
8. The system as described in claim 7 , wherein the portion alignment module further comprises a training data generation module to generate pairs of training digital images that include a training ground truth digital image and a training guidance digital image.
9. The system as described in claim 8 , wherein the training data generation module is configured to generate the training guidance digital image by:
selecting a first digital image as the training ground truth digital image and a second digital image;
adding a portion from the first digital image to the second digital image; and
blending the added portion as part of the second digital image.
10. The system as described in claim 8 , wherein the training data generation module is configured to further generate the training guidance digital image by:
selecting a training portion of the second digital image that encompasses the blended portion and a part of the second digital image in addition to the blended portion; and
generating the training guidance digital image using the selected training portion.
11. In a digital medium environment to fill a hole in a digital image using a guidance digital image, a method implemented by a computing device, the method comprising:
generating, by the computing device, parameters to align a portion of the guidance digital image with the hole to be filled in the digital image; and
filling, by the computing device, the hole in the digital image based at least in part on the parameters, the filling of the hole including:
identifying a first part of the hole in the digital image to be blended with the guidance digital image;
identifying a second part of the hole in the digital image, the second part of the hole separate from the first part of the hole;
blending the portion of the guidance digital image within the first part of the hole in the digital image; and
generating pixels for the second part of the hole in the digital image using a model trained using machine learning.
12. The method as described in claim 11 , wherein the parameters specify at least one of a translation, a rotation, or a scaling to be applied to align the portion of the guidance digital image with respect to the hole to be filled in the digital image.
13. The method as described in claim 11 , wherein the blending is performed by a model trained using machine learning.
14. The method as described in claim 11 , further comprising:
generating two or more loss function values based on an amount of realism exhibited by the filled digital image; and
weighting the two or more loss function values of the filled digital image, each loss function value weighted with respect to each other loss function value.
15. The method as described in claimer 14 , further comprising:
ranking the weighted loss function values of a plurality of said filled digital images.
16. The method as described in claim 11 , wherein the generated pixels are based on at least part of the digital image with the hole to be filled.
17. The method as described in claim 11 , wherein the generating of the parameters uses a model trained using machine learning.
18. The method as described in claim 17 , further comprising generating pairs of training digital images that include a training ground truth digital image and a training guidance digital image.
19. In a digital medium environment to fill a hole in a digital image using a guidance digital image, a system 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 including:
generating parameters to align a portion of the guidance digital image with the hole to be filled in the digital image;
identifying first and second parts of the hole using a model trained using machine learning; and
filling the hole in the digital image based at least in part on the parameters, the filling of the hole including:
blending the portion of the guidance digital image within the first part of the hole in the digital image; and
generating pixels using a model trained using machine learning to fill the second part of the hole in the digital image that is separate from the first part of the hole in the digital image.
20. The system as described in claim 19 , wherein the parameters specify at least one of a translation, a rotation, or a scaling to be applied to align the portion of the guidance digital image with respect to the hole to be filled in the digital image.