Granted Patent B2

US 12,340,603 · App. 18/000,737 · Granted Jun 24, 2025

Inventors: Yan Xiang (Shenzhen, CN); Xiao Zhang (Shenzhen, CN); Ke Xu (Shenzhen, CN); Fang Zhu (Shenzhen, CN)

Assignee: SANECHIPS TECHNOLOGY CO., LTD.

G06V20/70G06N3/045G06N3/0475G06V10/82

Abstract

Claims (36)

1. A method for labeling an object contour in a target image, comprising:

acquiring a target image feature of a target image, wherein the target image comprises a target object of a target type;

inputting the target image feature into a target generator, wherein the target generator is a generator in a generative adversarial network trained by utilizing a sample image, the generative adversarial network comprises the target generator and a discriminator, the target generator is configured to generate a first mask of the sample image upon acquiring a first image feature of the sample image, and the discriminator is configured to, upon receiving a sample image obtained after pixels corresponding to the first mask are erased, identify the type of a sample object in the sample image obtained after the pixels are erased, the type of the sample object is used for training parameters in the target generator; and

acquiring a target mask of the target image generated by the target generator, wherein the target mask is used for labeling a contour of the target object.

2. The method of claim 1 , prior to the inputting the target image feature into the target generator, further comprising:

acquiring a first image feature of the sample image;

inputting the first image feature into the target generator to generate the first mask of the sample image;

erasing pixels corresponding to the first mask to obtain a first image;

inputting the first image and the sample image into the discriminator to train the discriminator; and

inputting the first image into the target generator to train the target generator.

3. The method of claim 2 , wherein the inputting the first image and the sample image into the discriminator to train the discriminator comprises:

calculating a first loss of the discriminator after the first image and the sample image are input into the discriminator; and

adjusting parameters in the discriminator by utilizing the first loss.

4. The method of claim 2 , wherein the inputting the first image into the target generator to train the target generator comprises:

acquiring a first type of a first object in the first image output after the first image is input into the discriminator;

calculating a second loss of the target generator under the first type; and

adjusting parameters in the target generator by utilizing the second loss.

5. The method of claim 2 , wherein the acquiring a target image feature of a target image comprises:

acquiring the target image;

inputting the target image into a target model, the target model being a model obtained after deleting a fully-connected layer of a pre-trained first model; and

acquiring the target image feature of the target image output by the target model.

6. The method of claim 5 , prior to the inputting the target image feature into the target generator, further comprising:

acquiring the sample image;

training a second model by utilizing the sample image to obtain the trained first model, the second model being the first model before training; and

deleting a fully-connected layer of the first model to obtain the target model.

7. The method of claim 5 , wherein convolution layers of the discriminator and the first model comprise dilated convolutions with different dilation coefficients.

8. An apparatus for labeling an object contour in a target image, comprising:

a first acquisition unit, configured to acquire a target image feature of a target image, wherein the target image comprises a target object of a target type;

a first input unit, configured to input the target image feature into a target generator, wherein the target generator is a generator in a generative adversarial network trained by utilizing a sample image, the generative adversarial network comprises the target generator and a discriminator, the target generator is configured to generate a first mask of the sample image upon acquiring a first image feature of the sample image, and the discriminator being configured to, upon receiving a sample image obtained after pixels corresponding to the first mask are erased, identify the type of a sample object in the sample image obtained after the pixels are erased, the type of the sample object is used for training parameters in the target generator; and

a second acquisition unit, configured to acquire a target mask of the target image generated by the target generator, wherein the target mask is used for labeling a contour of the target object.

9. A non-transitory computer-readable storage medium having computer programs stored thereon which, when executed by a processor, cause the processor to carry out the method of claim 1 .

10. An electronic device, comprising a memory and a processor, the memory storing computer programs which, when executed by a processor, cause the processor to carry out a method for labeling an object contour in a target image, the method comprising:

acquiring a target image feature of a target image, wherein the target image comprises a target object of a target type;

inputting the target image feature into a target generator, wherein the target generator is a generator in a generative adversarial network trained by utilizing a sample image, the generative adversarial network comprises the target generator and a discriminator, the target generator is configured to generate a first mask of the sample image upon acquiring a first image feature of the sample image, and the discriminator is configured to, upon receiving a sample image obtained after pixels corresponding to the first mask are erased, identify the type of a sample object in the sample image obtained after the pixels are erased, the type of the sample object is used for training parameters in the target generator; and

acquiring a target mask of the target image generated by the target generator, wherein the target mask is used for labeling a contour of the target object.

11. The method of claim 6 , wherein convolution layers of the discriminator and the first model comprise dilated convolutions with different dilation coefficients.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (21)