Cross-modal shape and color manipulation

Patent metadata

Field	Value
Publication number	US-12094073-B2
Application number	US-202217814391-A
Country	US
Kind code	B2
Filing date	Jul 22, 2022
Priority date	May 31, 2022
Publication date	Sep 17, 2024
Grant date	Sep 17, 2024

Abstract

Systems, computer readable media, and methods herein describe an editing system where a three-dimensional (3D) object can be edited by editing a 2D sketch or 2D RGB views of the 3D object. The editing system uses multi-modal (MM) variational auto-decoders (VADs)(MM-VADs) that are trained with a shared latent space that enables editing 3D objects by editing 2D sketches of the 3D objects. The system determines a latent code that corresponds to an edited or sketched 2D sketch. The latent code is then used to generate a 3D object using the MM-VADs with the latent code as input. The latent space is divided into a latent space for shapes and a latent space for colors. The MM-VADs are trained with variational auto-encoders (VAE) and a ground truth.

First claim

What is claimed is:

1. A method comprising:
accessing a first two-dimensional (2D) sketch;
determining, by one or more processors, a first latent code corresponding to the first 2D sketch based on a first variational auto-decoder (VAD);
generating a second 2D sketch based on the first VAD with the first latent code as input;
causing the second 2D sketch to be displayed;
determining the second 2D sketch has been edited;
determining a second latent code corresponding to the edited 2D sketch based on the first VAD;
generating a three-dimensional (3D) shape from the second latent code based on a second VAD with the second latent code as input; and
causing to be displayed the 3D shape on the display of a computing device.

2. The method of claim 1 wherein determining the second latent code further comprises:
generating a third 2D sketch based on the first VAD with the second latent code as input;
determining a loss between the third 2D sketch and the second 2D sketch; and
in response to the loss being less than a threshold, determining to use the second latent code for the second 2D sketch.

3. The method of claim 2 wherein the second latent code comprises a color latent code and a shape latent code.

4. The method of claim 3 wherein the second 2D sketch is generated with only the shape latent code of the second latent code and the 3D shape is generated with both the color latent code and the shape latent code.

5. The method of claim 3 wherein the second VAD comprises a third VAD and fourth VAD, and wherein the generating the 3D shape further comprises:
generating a three-dimensional (3D) color based on the third VAD with the color latent code as input;
generating a signed distance field (SDF) based on the fourth VAD with the shape latent code as input; and
combining the 3D color and the SDF to generate the 3D shape.

6. The method of claim 2 wherein the threshold is a first threshold, the loss is a first loss, and wherein the method further comprises:
determining a loss between a previous latent code and the first latent code; and, wherein the in response to the loss further comprises:
in response to the first loss being less than the first threshold and the second loss being less than a second threshold, determining to use the first latent code for the first 2D sketch.

7. The method of claim 1 wherein the 3D shape is a first 3D shape and wherein the method further comprises:
generating a second 3D shape from the first latent code based on the second VAD with the first latent code as input; and
displaying the second 3D shape on a display of a computing device.

8. The method of claim 1 further comprising:
generating a 2D color view based on a third VAD with the first latent code and a view as input; and
displaying the 2D color view on the display of the computing device.

9. The method of claim 8 wherein the 2D color view is a first 2D color view, the 3D shape is a first 3D shape, and wherein the method further comprises:
determining the first 2D color view comprises edits;
determining a third latent code corresponding to the edited 2D color view based on the third VAD;
generating a second 2D color view based on the third VAD with the third latent code as input;
generating a third 2D sketch based on the first VAD with the third latent code as input;
generating a second 3D shape based on the third VAD with the third latent code as input; and
displaying the second 2D color view, the third 2D sketch, and the second 3D shape on the display of the computing device.

10. The method of claim 9 wherein the edits of the first 2D color view are edits that change the color of the first 2D color view.

11. The method of claim 1 wherein determining the first latent code further comprises:
determining a plurality of latent codes based on a loss between the first 2D sketch and a plurality of 2D sketches generated from the plurality of latent codes;
generating a plurality of 3D shapes based on the second VAD with the plurality of latent codes as inputs;
displaying the plurality of 3D shapes on the display of the computing device; and
in response to a selection of a 3D shape of the plurality of 3D shapes, determining a corresponding latent code of the plurality of latent codes used as input to generate the selected 3D shape is the first latent code.

12. The method of claim 1 wherein the method further comprises:
generating a third latent code based on a first variational auto-encoder (VAE) with a ground truth 2D sketch as input;
generating a third 2D sketch based on the first VAD with the third latent code as input; and
adjusting weights of the first VAD and the first VAE based on a difference between the third 2D sketch and the ground truth 2D sketch.

13. The method of claim 12 wherein the 3D shape is a first 3D shape and wherein the method further comprises:
generating a fourth latent code based on a second VAE with a ground truth 3D shape as input, wherein the ground truth 3D shape and the ground truth 2D sketch are a matched pair;
generating a second 3D shape based on the second VAD with the fourth latent code as input; and
adjusting weights of the second VAD and the second VAE based on a difference between the second 3D shape and the ground truth 3D shape.

14. The method of claim 1 wherein the first VAD and the second VAD are fully connected neural networks with three to eight layers and wherein the first VAD and the second VAD are trained based on matched pairs of 2D sketches and corresponding 3D shapes.

15. The method of claim 1 further comprising:
training the first VAD and the second VAD to learn a mapping between a latent space comprising the first latent code and a second latent space wherein latent codes of a plurality of 2D sketches depicting a same 3D shape map to a same area of the second latent space.

16. A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by one or more processors, cause the one or more processors to perform operations comprising:
accessing a first two-dimensional (2D) sketch;
determining, by one or more processors, a first latent code corresponding to the first 2D sketch based on a first variational auto-decoder (VAD);
generating a second 2D sketch based on the first VAD with the first latent code as input;
causing the second 2D sketch to be displayed;
determining the second 2D sketch has been edited;
determining a second latent code corresponding to the edited 2D sketch based on the first VAD;
generating a three-dimensional (3D) shape from the second latent code based on a second VAD with the second latent code as input; and
causing to be displayed the 3D shape on the display of a computing device.

17. The computer-readable storage medium of claim 16 , wherein the determining the second latent code further comprises:
generating a third 2D sketch based on the first VAD with the second latent code as input;
determining a loss between the third 2D sketch and the second 2D sketch; and
in response to the loss being less than a threshold, determining to use the second latent code for the second 2D sketch.

18. A system comprising:
one or more processors; and
a memory storing instructions that, when executed by the one or more processors, configure the system to perform operations comprising:
accessing a first two-dimensional (2D) sketch;
determining, by one or more processors, a first latent code corresponding to the first 2D sketch based on a first variational auto-decoder (VAD);

Assignees

• Snap Inc

Inventors

• Chai Menglei
• Tulyakov Sergey
• Ren Jian
• Lee Hsin-Ying
• Olszewski Kyle
• Huang Zeng
• Cheng Zezhou

Classifications

• G06T17/00

  Three-dimensional [3D] modelling for computer graphics · CPC title

• G06T2219/2012

  Colour editing, changing, or manipulating; Use of colour codes · CPC title

• G06T2219/2021

  Shape modification · CPC title

• G06T19/20 Primary

  Editing of three-dimensional [3D] images, e.g. changing shapes or colours, aligning objects or positioning parts · CPC title