Granted Patent B1

US 12,198,430 · App. 17/009,542 · Granted Jan 14, 2025

Inventor: Satwik Kottur (Menlo Park, CA)

Assignee: Meta Platforms, Inc.

G06V20/35G06F16/9024G06F16/9536G06V20/20

Abstract

Claims (80)

1. A method comprising:

receiving, from a client system associated with a user, a first user request comprising a reference to a target object and one or more of an attribute or a relationship of the target object;

accessing, from the client system, visual data captured by one or more cameras of the client system, wherein the visual data comprises images portraying the target object;

resolving, based on a multimodal dialog state, the reference to the target object portrayed in the images captured by the one or more cameras of the client system;

determining that first object information, corresponding to the attribute or the relationship, is not already stored in the multimodal dialog state;

responsive to resolving the reference to the target object and the first object information not being stored in the multimodal dialog state,

executing a first visual analysis by retrieving one or more of the images, and identifying the first object information from the one or more of the images based on an identification of the target object in the one or more of the images and by analyzing the one or more of the images, and

halting provisioning of other images of the images in response to the first object information being identified from the one or more of the images; and

storing, responsive to receiving the first user request, the first object information of the target object as an incremental update in the multimodal dialog state.

2. The method of claim 1 , further comprising:

sending, to the client system, instructions for presenting a response to the first user request, wherein the response comprises the first object information.

3. The method of claim 1 , wherein the first object information is stored as a scene graph in the multimodal dialog state.

4. The method of claim 1 , further comprising:

receiving, from the client system, the visual data;

analyzing, by a computer vision module, the visual data to identify one or more objects portrayed in the one or more images captured by the one or more cameras of the client system; and

assigning respective object identifiers to one or more of the identified one or more objects.

5. The method of claim 4 , wherein resolving the reference to the target object comprises:

selecting, from among the respective object identifiers of the one or more identified objects, a target object identifier corresponding to the target object.

6. The method of claim 5 , wherein determining the first object information of the target object comprises:

providing, in response to receiving the first user request, the visual data and the target object identifier of the target object to a scene understanding engine; and

generating, by the scene understanding engine, the first object information based on the target object identifier.

7. The method of claim 6 , further comprising:

generating, by a dialog engine, a partial scene graph consisting of the target object identifier and the first object information of the target object;

receiving, from the client system, a second user request comprising another reference to the target object;

resolving, based on the multimodal dialog state, the another reference to the target object, wherein a second object information about the target object is not already stored in the multimodal dialog state; and

responsive to resolving the another reference to the target object, executing a second visual analysis of the one or more of the images to determine the second object information, and storing the second object information as an incremental update to the partial scene graph,

wherein storing the first object information of the target object comprises storing the partial scene graph in the multimodal dialog state.

8. The method of claim 6 , further comprising:

in response to the generation of the first object information, halting provision of the visual data to the scene understanding engine.

9. The method of claim 1 , wherein resolving the reference to the target object comprises:

generating, by a dialog engine, an initial scene graph comprising the target object and one or more additional objects portrayed in the one or more images captured by the one or more cameras of the client system; and

resolving the reference to the target object based on the initial scene graph.

10. The method of claim 9 , further comprising:

selecting a portion of the initial scene graph comprising the target object and the first object information of the target object from among a plurality of portions of the initial scene graph, wherein remaining portions of the initial scene graph do not comprise the target object and the first object information.

11. The method of claim 10 , wherein storing the first object information of the target object comprises:

storing the selected portion of the initial scene graph in the multimodal dialog state; and

deleting the remaining portions of the initial scene graph.

12. The method of claim 1 , further comprising:

determining, by a scene understanding engine, one or more properties of the target object based on the first visual analysis of the one or more images captured by the one or more cameras of the client system; and

resolving the target object to a specific entity based on the one or more properties.

13. The method of claim 12 , further comprising:

accessing a knowledge graph based on the specific entity; and

retrieving the first object information from the knowledge graph.

14. The method of claim 1 , further comprising:

receiving, from the client system, a second user request comprising a reference to the target object and one or more of an additional attribute or an additional relationship of the target object;

determining a second object information of the target object corresponding to the referenced additional attribute or relationship of the second user request based on a subsequent visual analysis of the one or more images captured by the one or more cameras of the client system; and

storing, responsive to receiving the second user request, the second object information of the target object in the multimodal dialog state.

15. The method of claim 1 , further comprising:

receiving additional visual data, the additional visual data comprising one or more additional images captured by the one or more cameras of the client system portraying a second object sharing an attribute or relationship with the target object.

16. The method of claim 15 , further comprising:

determining whether a second user request referencing the second object has been received; and

responsive to determining whether the second user request referencing the second object has been received:

if the second user request has been received, storing, responsive to receiving the second user request, a second object information of the second object in the multimodal dialog state; and

if the second user request has not been received, storing the additional visual data without storing the second object information of the second object in the multimodal dialog state.

17. The method of claim 15 , further comprising:

receiving, from the client system, a second user request, wherein the second user request comprises a reference to the shared attribute or relationship; and

determining, based on second object information of the second object stored in the multimodal dialog state, that the second user request is associated with the target object.

18. The method of claim 17 , further comprising:

sending, to the client system and in response to the second user request, instructions for presenting a response to the second user request, wherein the response comprises the second object information of the target object.

19. The method of claim 1 , wherein the first user request is received during a current dialog session between the user and an assistant system associated with the client system, and wherein the first object information of the target object is stored during the current dialog session.

20. One or more computer-readable non-transitory storage media embodying software that is operable when executed to:

receive, from a client system associated with a user, a first user request comprising a reference to a target object and one or more of an attribute or a relationship of the target object;

access, from the client system, visual data captured by one or more cameras of the client system, wherein the visual data comprises images portraying the target object;

resolve, based on a multimodal dialog state, the reference to the target object portrayed in the images captured by the one or more cameras of the client system;

determining that first object information, corresponding to the attribute or the relationship, is not already stored in the multimodal dialog state;

responsive to resolving the reference to the target object and the first object information not being stored in the multimodal dialog state,

executing a visual analysis by retrieving one or more of the images, and identifying the first object information from the one or more of the images based on an identification of the target object in the one or more of the images and by analyzing the one or more of the images, and

halting provisioning of other images of the images in response to the first object information being identified from the one or more of the images; and

store, responsive to receiving the first user request, the first object information of the target object as an incremental update in the multimodal dialog state.

21. A system comprising:

one or more processors; and

a non-transitory memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:

receive, from a client system associated with a user, a first user request comprising a reference to a target object and one or more of an attribute or a relationship of the target object;

access, from the client system, visual data captured by one or more cameras of the client system, wherein the visual data comprises images portraying the target object;

resolve, based on a multimodal dialog state, the reference to the target object portrayed in the images captured by the one or more cameras of the client system;

determining that first object information, corresponding to the attribute or the relationship, is not already stored in the multimodal dialog state;

responsive to resolving the reference to the target object and the first object information not being stored in the multimodal dialog state,

executing a visual analysis by retrieving one or more of the images, and identifying the first object information from the one or more of the images based on an identification of the target object in the one or more of the images and by analyzing the one or more of the images, and

halting provisioning of other images of the images in response to the first object information being identified from the one or more of the images; and

store, responsive to receiving the first user request, the first object information of the target object as an incremental update in the multimodal dialog state.

Assignments (2)

References Cited (210)

Cited By (1)

US 12,400,289