Semantically-guided template generation from image content

1. A method for generating a template from an input banner image, the method comprising:

receiving, from a user via a user interface, an indication of the input banner image to be emulated, the input banner image having an arrangement of contents including images and textual elements;

extracting, using a first stage of a single processing pipeline, layout information of the input banner image, the layout information including a position and a content type of each of the contents in the input banner image; providing the layout information to a second stage of the single processing pipeline;

extracting, using the second stage of the single processing pipeline, text attributes from the textual elements of the input banner image by superimposing each region of textual elements on a plain background to form modified regions of textual elements and processing the modified regions of textual elements in parallel, wherein the text attributes of a textual element includes at least a font of text used in the textual element, and wherein each modified region of textual elements has been modified at least by superimposing the corresponding region of textual elements on the plain background to correspond to training data used by a model in the second stage of the single processing pipeline;

generating, using a third stage of the single processing pipeline, the template having an arrangement of editable regions corresponding to the arrangement of the contents of the input banner image; and

modifying the template by tagging the textual elements of the editable regions with corresponding text attributes from the input banner image.

2. The method of claim 1 , wherein the content type of a content in the input banner image is selected from a set of content types including an image content type, a text content type, and a shape content type.

3. The method of claim 1 , wherein the template is stored in a data structure of a data exchange format indicating layout information of the editable regions.

4. The method of claim 1 , wherein the text attributes further indicate at least one of: a color of the text, a size of the text, or a formatting of the text used in the textual element.

5. The method of claim 1 , wherein the input banner image is a flat image that lacks content layer information.

6. The method of claim 1 , further comprising identifying, using a fourth stage of the single processing pipeline, a position of a logo included in the input banner image, and modifying the template by adding an editable logo region corresponding to the position of the logo in the input banner image.

7. A non-transitory computer-readable medium comprising instructions that, when executed by at least one processor, cause the processor to perform operations to generate a template from an input banner image, the operations comprising:

receiving, from a user via a user interface, an indication of the input banner image to be emulated, the input banner image having an arrangement of contents including images and textual elements;

extracting, using a first stage of a single processing pipeline, layout information of the input banner image, the layout information including a position and a content type of each of the contents in the input banner image; providing the layout information to a second stage of the single processing pipeline;

extracting, using the second stage of the single processing pipeline, text attributes from the textual elements of the input banner image by superimposing each region of textual elements on a plain background to form modified regions of textual elements and processing the modified regions of textual elements in parallel, wherein the text attributes of a textual element includes at least a font of text used in the textual element, and wherein each modified region of textual elements has been modified at least by superimposing the corresponding region of textual elements on the plain background to correspond to training data used by a model in the second stage of the single processing pipeline;

generating, using a third stage of the single processing pipeline, the template having an arrangement of editable regions corresponding to the arrangement of the contents of the input banner image; and

modifying the template by tagging the textual elements of the editable regions with corresponding text attributes from the input banner image.

8. The non-transitory computer-readable medium of claim 7 , wherein the content type of a content in the input banner image is selected from a set of content types including an image content type, a text content type, and a shape content type.

9. The non-transitory computer-readable medium of claim 7 , wherein the template is stored in a data structure of a data exchange format indicating layout information of the editable regions.

10. The non-transitory computer-readable medium of claim 7 , wherein the text attributes further indicate at least one of: a color of the text, a size of the text, or a formatting of the text used in the textual element.

11. The non-transitory computer-readable medium of claim 7 , wherein the input banner image is a flat image that lacks content layer information.

12. The non-transitory computer-readable medium of claim 7 , wherein the operations further comprise identifying, using a fourth stage of the single processing pipeline, a position of a logo included in the input banner image, and modifying the template by adding an editable logo region corresponding to the position of the logo in the input banner image.

13. A computing system for generating a template from an input banner image, comprising:

one or more processors; and

a computer-readable storage medium, coupled with the one or more processors, having instructions stored thereon, which, when executed by the one or more processors, cause the one or more processors to perform operations comprising:

receiving, from a user via a user interface, an indication of the input banner image to be emulated, the input banner image having an arrangement of contents including images and textual elements;

extracting, using a first stage of a single processing pipeline, layout information of the input banner image, the layout information including a position and a content type of each of the contents in the input banner image; providing the layout information to a second stage of the single processing pipeline;

extracting, using the second stage of the single processing pipeline, text attributes from the textual elements of the input banner image by superimposing each region of textual elements on a plain background to form modified regions of textual elements and processing the modified regions of textual elements in parallel, wherein the text attributes of a textual element includes at least a font of text used in the textual element, and wherein each modified region of textual elements has been modified at least by superimposing the corresponding region of textual elements on the plain background to correspond to training data used by a model in the second stage of the single processing pipeline;

generating, using a third stage of the single processing pipeline, the template having an arrangement of editable regions corresponding to the arrangement of the contents of the input banner image; and

modifying the template by tagging the textual elements of the editable regions with corresponding text attributes from the input banner image.

14. The computing system of claim 13 , wherein the content type of a content in the input banner image is selected from a set of content types including an image content type, a text content type, and a shape content type.

15. The computing system of claim 13 , wherein the template is stored in a data structure of a data exchange format indicating layout information of the editable regions.

16. The computing system of claim 13 , wherein the text attributes further indicate at least one of: a color of the text, a size of the text, or a formatting of the text used in the textual element.

17. The computing system of claim 13 , wherein the input banner image is a flat image that lacks content layer information.

18. The computing system of claim 13 , wherein the operations further comprise identifying, using a fourth stage of the single processing pipeline, a position of a logo included in the input banner image, and modifying the template by adding an editable logo region corresponding to the position of the logo in the input banner image.