# US 12,112,402 B2 · Method, electronic device, and computer program product for processing target object

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

- Number: US 12,112,402 B2
- Document ID: US12112402B2
- Status: Granted
- Kind: B2
- Application number: 17/969,782
- Filed: 2022-10-20
- Published: 2024-03-21
- Granted: 2024-10-08
- Earliest claimed: 2022-09-16
- Assignees: Dell Products L.P.
- Applicants: Dell Products L.P.
- Inventors: Zijia Wang; Zhisong Liu; Zhen Jia
- Web page: https://patentgrep.com/patent/US12112402B2

## CPC

- G06T 11/00: Two-dimensional [2D] image generation
- G06F 40/20: Natural language analysis (semantic analysis of natural language G06F40/30)
- G06F 40/30: Semantic analysis
- G06T 1/0021: {Image watermarking}
- G10L 15/02: Feature extraction for speech recognition; Selection of recognition unit
- G10L 25/06: the extracted parameters being correlation coefficients
- G10L 25/57: for processing of video signals
- G10L 25/63: for estimating an emotional state
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for processing a target object. The method includes acquiring an initial non-video feature vector on the basis of at least one input of a received speech input and text input. The method further includes taking, in response to not receiving a video input, a default feature vector as an initial video feature vector corresponding to the video input. The method further includes generating a video feature, a speech feature, and a text feature on the basis of the initial non-video feature vector and the initial video feature vector. The method further includes generating a processing parameter for a target object on the basis of the video feature, the speech feature, and the text feature, wherein the processing parameter includes at least one of an emotion parameter, an attribute parameter, and a pose parameter.

## Claims

### Claim 1

A method, comprising:
acquiring an initial non-video feature vector on the basis of at least one input of a received speech input and text input;
taking, in response to not receiving a video input, a default feature vector as an initial video feature vector in place of the video input;
generating a video feature, a speech feature, and a text feature on the basis of the initial non-video feature vector and the initial video feature vector; and
generating a processing parameter for a target object on the basis of the video feature, the speech feature, and the text feature, wherein the processing parameter includes at least one of an emotion parameter, an attribute parameter, and a pose parameter;
wherein generating the processing parameter for the target object comprises generating the processing parameter utilizing a trained parameter generator model; and
wherein the parameter generator model is trained over a plurality of iterations of an iterative training process in which the parameter generator model is adjusted based on distances between at least portions of respective distinct training parameter sets obtained in respective different ones of the iterations of the iterative training process.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the initial non-video feature vector comprises at least one of an initial speech feature vector and an initial text feature vector corresponding to the at least one input.

### Claim 3 (depends on 2)

The method according to claim 2, wherein the default feature vector comprises an all-zero vector, and the taking a default feature vector as an initial video feature vector in place of the video input comprises:
acquiring the all-zero vector as the initial video feature vector in response to not receiving the video input within a predetermined time period.

### Claim 4 (depends on 2)

The method according to claim 2, further comprising:
setting the initial text feature vector corresponding to the text input to be an all-zero vector in response to only receiving the speech input.

### Claim 5 (depends on 2)

The method according to claim 2, further comprising:
setting the initial speech feature vector corresponding to the speech input to be an all-zero vector in response to only receiving the text input.

### Claim 6 (depends on 2)

The method according to claim 2, wherein the generating the video feature, the speech feature, and the text feature on the basis of the initial video feature vector and the initial non-video feature vector comprises:
generating an indication of a correlation among the initial video feature vector, the initial speech feature vector, and the initial text feature vector; and
generating the video feature, the speech feature, and the text feature on the basis of the indication of the correlation.

### Claim 7 (depends on 6)

The method according to claim 6, wherein the indication of the correlation comprises a synthetic tensor, and the generating the video feature, the speech feature, and the text feature on the basis of the indication of the correlation comprises:
decomposing the synthetic tensor to generate the video feature, the speech feature, and the text feature for the target object.

### Claim 8 (depends on 6)

The method according to claim 6, wherein the indication of the correlation represents the correlation among the initial video feature vector, the initial speech feature vector, and the initial text feature vector.

### Claim 9 (depends on 1)

The method according to claim 1, wherein the parameter generator model is trained at least in part by:
in a first round of iteration, respectively inputting each of a sample video input, a sample speech input, and a sample text input and/or any combination of the sample video input, the sample speech input, and the sample text input into the parameter generator model to obtain a corresponding first training parameter set;
in a second round of iteration, respectively inputting each of the sample video input, the sample speech input, and the sample text input and/or any combination of the sample video input, the sample speech input, and the sample text input into the parameter generator model to obtain a corresponding second training parameter set;
determining a distance between the first training parameter set and the second training parameter set; and
adjusting the parameter generator model on the basis of the determined distance to obtain the trained parameter generator model.

### Claim 10 (depends on 1)

The method according to claim 1, further comprising:
generating random noise; and
adding the random noise to an image to be processed that comprises the target object, so as to add a watermark to the image to be processed.

### Claim 11 (depends on 1)

The method according to claim 1, further comprising:
rendering the target object according to the processing parameter;
generating a video that comprises the rendered target object; and
outputting the generated video.

### Claim 12

An electronic device, comprising:
at least one processor; and
a memory coupled to the at least one processor and having instructions stored thereon, wherein the instructions, when executed by the at least one processor, cause the electronic device to perform actions comprising:
acquiring an initial non-video feature vector on the basis of at least one input of a received speech input and text input;
taking, in response to not receiving a video input, a default feature vector as an initial video feature vector in place of the video input;
generating a video feature, a speech feature, and a text feature on the basis of the initial non-video feature vector and the initial video feature vector; and
generating a processing parameter for a target object on the basis of the video feature, the speech feature, and the text feature, wherein the processing parameter includes at least one of an emotion parameter, an attribute parameter, and a pose parameter;
wherein generating the processing parameter for the target object comprises generating the processing parameter utilizing a trained parameter generator model; and
wherein the parameter generator model is trained over a plurality of iterations of an iterative training process in which the parameter generator model is adjusted based on distances between at least portions of respective distinct training parameter sets obtained in respective different ones of the iterations of the iterative training process.

### Claim 13 (depends on 12)

The electronic device according to claim 12, wherein the initial non-video feature vector comprises at least one of an initial speech feature vector and an initial text feature vector corresponding to the at least one input.

### Claim 14 (depends on 13)

The electronic device according to claim 13, wherein the default feature vector comprises an all-zero vector, and the taking a default feature vector as an initial video feature vector in place of the video input comprises:
acquiring the all-zero vector as the initial video feature vector in response to not receiving the video input within a predetermined time period.

### Claim 15 (depends on 13)

The electronic device according to claim 13, wherein the actions further comprise:
setting the initial text feature vector corresponding to the text input to be an all-zero vector in response to only receiving the speech input.

### Claim 16 (depends on 13)

The electronic device according to claim 13, wherein the actions further comprise:
setting the initial speech feature vector corresponding to the speech input to be an all-zero vector in response to only receiving the text input.

### Claim 17 (depends on 13)

The electronic device according to claim 13, wherein the generating the video feature, the speech feature, and the text feature on the basis of the initial video feature vector and the initial non-video feature vector comprises:
generating an indication of a correlation among the initial video feature vector, the initial speech feature vector, and the initial text feature vector, wherein the indication of the correlation comprises a synthetic tensor; and
decomposing the synthetic tensor to generate the video feature, the speech feature, and the text feature for the target object.

### Claim 18 (depends on 12)

The electronic device according to claim 12, wherein the parameter generator model is trained at least in part by:
in a first round of iteration, respectively inputting each of a sample video input, a sample speech input, and a sample text input and/or any combination of the sample video input, the sample speech input, and the sample text input into the parameter generator model to obtain a corresponding first training parameter set;
in a second round of iteration, respectively inputting each of the sample video input, the sample speech input, and the sample text input and/or any combination of the sample video input, the sample speech input, and the sample text input into the parameter generator model to obtain a corresponding second training parameter set;
determining a distance between the first training parameter set and the second training parameter set; and
adjusting the parameter generator model on the basis of the determined distance to obtain the trained parameter generator model.

### Claim 19 (depends on 12)

The electronic device according to claim 12, wherein the actions further comprise:
generating random noise; and
adding the random noise to an image to be processed that comprises the target object, so as to add a watermark to the image to be processed.

### Claim 20

A computer program product that is tangibly stored on a non-transitory computer-readable medium and comprises machine-executable instructions, wherein the machine-executable instructions, when executed by a machine, cause the machine to perform actions comprising:
acquiring an initial non-video feature vector on the basis of at least one input of a received speech input and text input;
taking, in response to not receiving a video input, a default feature vector as an initial video feature vector in place of the video input;
generating a video feature, a speech feature, and a text feature on the basis of the initial non-video feature vector and the initial video feature vector; and
generating a processing parameter for a target object on the basis of the video feature, the speech feature, and the text feature, wherein the processing parameter includes at least one of an emotion parameter, an attribute parameter, and a pose parameter;
wherein generating the processing parameter for the target object comprises generating the processing parameter utilizing a trained parameter generator model; and
wherein the parameter generator model is trained over a plurality of iterations of an iterative training process in which the parameter generator model is adjusted based on distances between at least portions of respective distinct training parameter sets obtained in respective different ones of the iterations of the iterative training process.