# US 2021/0264649 A1 · Generative Shape Creation and Editing

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0264649 A1
- Document ID: US20210264649A1
- Status: Published application
- Kind: A1
- Application number: 17/317,246
- Filed: 2021-05-11
- Published: 2021-08-26
- Earliest claimed: 2020-01-15
- Assignees: Adobe Inc.
- Applicants: Adobe Inc.
- Inventors: Giorgio Gori; Tamy Boubekeur; Radomir Mech; Nathan Aaron Carr; Matheus Abrantes Gadelha; Duygu Ceylan Aksit
- Web page: https://patentgrep.com/patent/US20210264649A1

## CPC

- G06T 11/60: Creating or editing images; Combining images with text
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0475: Generative networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/00: Machine learning
- G06T 9/00: Image coding (bandwidth or redundancy reduction for static pictures H04N1/41; coding or decoding of static colour picture signals H04N1/64; methods or arrangements for coding, decoding, compressing or decompressing digital video signals H04N19/00)
- G06T 11/23: using straight lines or curves
- G06T 2200/24: involving graphical user interfaces [GUIs]
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Generative shape creation and editing is leveraged in a digital medium environment. An object editor system represents a set of training shapes as sets of visual elements known as “handles,” and converts sets of handles into signed distance field (SDF) representations. A handle processor model is then trained using the SDF representations to enable the handle processor model to generate new shapes that reflect salient visual features of the training shapes. The trained handle processor model, for instance, generates new sets of handles based on salient visual features learned from the training handle set. Thus, utilizing the described techniques, accurate characterizations of a set of shapes can be learned and used to generate new shapes. Further, generated shapes can be edited and transformed in different ways.

## Claims

### Claim 1

A method for generating shapes implemented by at least one computing device, the method comprising:
encoding, by the at least one computing device, an first set of visual elements to generate a latent representation;
decoding, by the at least one computing device, the latent representation to generate a set of decoded visual elements;
calculating, by the at least one computing device, an accuracy probability for individual decoded visual elements of the set of decoded visual elements;
removing, by that at least one computing device, a particular decoded visual element based on determining that an accuracy probability for the particular decoded visual element is below a threshold accuracy probability; and
generating, by the at least one computing device, a shape utilizing remaining decoded visual elements.

### Claim 2 (depends on 1)

A method as described in claim 1, further comprising transforming the first set of visual elements into distance field representations, and wherein said encoding comprises encoding the distance field representations to generate the latent representation.

### Claim 3 (depends on 1)

A method as described in claim 1, wherein said decoding comprises decoding the latent representation based on a complexity variable to generate the set of decoded visual elements to include more visual elements than the first set of visual elements.

### Claim 4 (depends on 3)

A method as described in claim 3, wherein the complexity variable is user specified.

### Claim 5 (depends on 3)

A method as described in claim 3, wherein the complexity variable specifies a number of visual elements to be included in the decoded visual elements.

### Claim 6 (depends on 1)

A method as described in claim 1, wherein said calculating an accuracy probability for individual decoded visual elements is performed by a machine learning model and comprises configuring a bidirectional chamfer distance of a loss function to enable the machine learning model to calculate the accuracy probability based on an effect of individual decoded visual elements on evaluation of the loss function.

### Claim 7 (depends on 1)

A method as described in claim 1, wherein said decoding is performed by a decoder module that is trained using the first set of visual elements to generate the set of decoded visual elements, and wherein said calculating an accuracy probability is performed by an accuracy predictor module that is trained using the set of decoded visual elements to perform said calculating the accuracy probability.

### Claim 8 (depends on 1)

A method as described in claim 1, further comprising:
receiving an input interaction to manipulate a visual element of the generated shape;
processing the manipulated visual element to adjust the manipulated visual element to a defined shape manifold for the generated shape, wherein the shape manifold is based on the latent representation; and
generating an edited shape based on the adjustment to the manipulated visual element.

### Claim 9

A method for generating shapes implemented by at least one computing device, the method comprising:
encoding, by the at least one computing device, a set of visual elements to generate a latent representation;
decoding, by the at least one computing device, the latent representation to generate a set of decoded visual elements;
generating, by the at least one computing device, a shape utilizing the decoded visual elements;
receiving an input interaction to manipulate a visual element of the generated shape;
processing the manipulated visual element to adjust the manipulated visual element to a defined shape manifold for the generated shape, wherein the shape manifold is based on the latent representation; and
generating an edited shape based on the adjustment to the manipulated visual element.

### Claim 10 (depends on 9)

A method as described in claim 9, further comprising transforming the set of visual elements into distance field representations, and wherein said encoding comprises encoding the distance field representations to generate the latent representation.

### Claim 11 (depends on 9)

A method as described in claim 9, wherein said decoding comprises decoding the latent representation based on a complexity variable to generate the set of decoded visual elements to include more visual elements than the set of visual elements.

### Claim 12 (depends on 11)

A method as described in claim 11, wherein said receiving an input interaction to manipulate a visual element is based on a user input to a graphical user interface, and where the complexity variable is determined based on a different user input to the graphical user interface to specify the complexity variable.

### Claim 13 (depends on 9)

A method as described in claim 9, further comprising:
calculating, by the at least one computing device, an accuracy probability for individual decoded visual elements of the set of decoded visual elements; and
removing, by that at least one computing device, a particular decoded visual element based on determining that an accuracy probability for the particular decoded visual element is below a threshold accuracy probability, wherein said generating a shape utilizes remaining decoded visual elements.

### Claim 14 (depends on 9)

A method as described in claim 9, wherein the shape manifold comprises a topological space that defines different configurations of decoded visual elements according to the latent representation.

### Claim 15 (depends on 9)

A method as described in claim 9, wherein said processing the manipulated visual element to adjust the manipulated visual element comprises one or more of automatically repositioning or resizing the visual element to conform to the latent representation.

### Claim 16

A system for generating shapes, the system comprising:
an encoder module implemented at least partially in hardware of at least one computing device to encode an first set of visual elements to generate a latent representation;
a decoder module implemented at least partially in the hardware of the at least one computing device to decode the latent representation based on a complexity variable to generate a set of decoded visual elements that includes more visual elements than the first set of visual elements; and
an object editor module implemented at least partially in the hardware of the at least one computing device to generate a shape utilizing the set of decoded visual elements.

### Claim 17 (depends on 16)

A system as described in claim 16, wherein the decoder module is implemented to receive the complexity variable as a value received via user input.

### Claim 18 (depends on 16)

A system as described in claim 16, further comprising an accuracy predictor module implemented at least partially in the hardware of the at least one computing device to:
calculate an accuracy probability for individual decoded visual elements of the set of decoded visual elements; and
remove a particular decoded visual element based on determining that an accuracy probability for the particular decoded visual element is below a threshold accuracy probability.

### Claim 19 (depends on 18)

A system as described in claim 18, wherein the decoder module is trained using the first set of visual elements to generate the set of decoded visual elements, and wherein the accuracy predictor module is trained using the set of decoded visual elements to calculate the accuracy probability.

### Claim 20 (depends on 16)

A system as described in claim 16, further comprising an object editor module implemented at least partially in the hardware of the at least one computing device to receive an input interaction to manipulate a visual element of the generated shape, and wherein the decoder module is further implemented to adjust the manipulated visual element to a defined shape manifold for the generated shape, wherein the shape manifold is based on the latent representation.
