Multi-person human body model reconstruction method for low-resolution image
A low-resolution image and high-resolution image technology, applied in the field of 3D vision, can solve the problem of relative accuracy, and achieve the effect of improving the amount of feature information and optimizing feature extraction.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0049] A multi-person human body model reconstruction method for low-resolution images, comprising the following steps:
[0050] S1. Perform data preprocessing on the public data set, reduce the resolution by downsampling the high-resolution image to obtain a low-resolution image, and scale the low-resolution image to a uniform size on the basis of maintaining the original aspect ratio of the image for training low-resolution branch networks;
[0051] The preprocessing process described in S1 mainly includes the following steps:
[0052] S101, maintaining the original aspect ratio of the high-resolution image, and down-sampling the high-resolution image to (208, 128) to obtain a low-resolution image;
[0053] S102, the low-resolution image is unified to (832, 512) by bilinear interpolation, and the insufficient part is filled with 0;
[0054] S2. Train a high-resolution branch network through the original high-resolution image, and the high-resolution branch network is divid...
Embodiment 2
[0078] see Figure 1-5 , based on a low-resolution image-oriented multi-person human body model reconstruction method described in Embodiment 1, the specific implementation process is as follows:
[0079] (1) Data preprocessing:
[0080] In the present invention, the published Human3.6M, MPI-INF 3DHP, COCO, MPII datasets are used, and the above datasets include crowd activities in various situations; in order to obtain low-resolution images, the original aspect ratio of the images is maintained On the basis of , the data is first downsampled to (208, 128), and the image is fixed to (832, 512) by bilinear interpolation for unified training, and the insufficient area is filled with 0;
[0081] (2) Dual-branch multi-person reconstruction network:
[0082] In the training process, the high-resolution image is first input into the high-resolution branch for training, and the feature information from the high-resolution image is obtained; then the network parameters of the branch ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


