Contextual instance decoupling-based multi-person pose estimation method and apparatus

1. A contextual instance decoupling (CID)-based multi-person pose estimation (MPPE) method, comprising:

acquiring a preset number of images containing multiple persons;

inputting the images containing multiple persons, as a training sample, into a CID-based MPPE model for training; and

performing pose estimation on a target image using the trained CID-based MPPE model,

the CID-based MPPE model being provided with an instance information abstraction module, a global feature decoupling module and a heatmap estimation module.

2. The CID-based MPPE method according to claim 1 , wherein the CID-based MPPE model further comprises a backbone network, and the performing pose estimation on a target image using the trained CID-based MPPE model comprises:

inputting the target image into the backbone network to obtain a global feature map, the target image comprising multiple persons and the global feature map containing three-dimensional features of all the persons;

inputting the global feature map into the instance information abstraction module and the global feature decoupling module respectively;

obtaining an instance feature of each person in the target image through the instance information abstraction module;

inputting the instance feature of each person into the global feature decoupling module, and decoupling, by the global feature decoupling module, an instance feature perception map based on the global feature map and the instance feature of each person; and

inputting the instance feature perception map into the heatmap estimation module to obtain a probability distribution of each keypoint of each person in the target image.

3. The CID-based MPPE method according to claim 2 , wherein the obtaining an instance feature of each person in the target image through the instance information abstraction module comprises:

inputting the global feature map into a heatmap module;

extracting center point coordinates of each person; and

performing sampling according to corresponding locations of center point coordinates of each person in the global feature map to obtain the instance feature of each person in the target image.

4. The CID-based MPPE method according to claim 3 , wherein before the obtaining an instance feature of each person in the target image, the method further comprises:

recalibrating a center point feature of each person based on a spatial attention or a channel attention.

5. The CID-based MPPE method according to claim 2 , wherein the decoupling, by the global feature decoupling module, an instance feature perception map based on the global feature map and the instance feature of each person comprises:

recalibrating the instance feature of each person from a spatial dimension based on a mapping relationship between the instance feature of each person and the global feature map to obtain a first instance feature perception map;

recalibrating the instance feature of each person from a channel dimension based on a mapping relationship between the instance feature of each person and the global feature map to obtain a second instance feature perception map; and

fusing the first instance feature perception map and the second instance feature perception map to obtain the instance feature perception map.

6. The CID-based MPPE method according to claim 5 , wherein the recalibrating the instance feature of each person from a spatial dimension based on a mapping relationship between the instance feature of each person and the global feature map to obtain a first instance feature perception map comprises:

generating a spatial mask for each person in the global feature map to represent a weight of a foreground feature of each person; and

increasing the weight of the foreground feature, and recalibrating a spatial location in the instance feature of each person to obtain the first instance feature perception map.

7. The CID-based MPPE method according to claim 6 , wherein decoupling different channels comprises: re-weighting the global feature map over the channel dimension based on person features to generate the second instance feature perception map.

8. The CID-based MPPE method according to claim 7 , wherein the fusing the first instance feature perception map and the second instance feature perception map to obtain the instance feature perception map comprises: performing a weighted sum on the first instance feature perception map and the second instance feature perception map to obtain the instance feature perception map.

9. The CID-based MPPE method according to claim 2 , wherein the inputting the instance feature perception map into the heatmap estimation module to obtain a probability distribution of each keypoint of each person in the target image comprises:

inputting the instance feature perception map into the heatmap estimation module to obtain a heatmap corresponding to each person in the target image,

the heatmap corresponding to each person in the target image containing a probability distribution of each keypoint.

10. The CID-based MPPE method according to claim 1 , wherein the training in the CID-based MPPE model comprises training the CID-based MPPE model through a preset loss function:

=+λ GFD

where IIA represents a loss of the instance information abstraction module, GFD represents a loss of the global feature decoupling module, and λ represents a weight.

11. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method according to claim 1 .

12. A computer program product, comprising a computer program which, when executed by a processor, implements the steps of the method according to claim 1 .

13. A contextual instance decoupling (CID)-based multi-person pose estimation (MPPE) apparatus, comprising:

an acquisition module, configured to acquire a preset number of images containing multiple persons;

a training module, configured to input the images containing multiple persons, as a training sample, into a CID-based MPPE model for training; and

a pose estimation module, configured to perform pose estimation on a target image using the trained CID-based MPPE model,

the CID-based MPPE model being provided with an instance information abstraction module, a global feature decoupling module and a heatmap estimation module.