Shulou(Shulou.com)06/01 Report--

This article introduces how to use AI to generate 3D high-resolution reconstruction of adults from 2D images in PIFuHD. The content is very detailed. Interested friends can use it for reference. I hope it will be helpful to you.

First of all, let's see how they do it.

Researchers at Facebook and the University of Southern California recently published a new paper called PIFuHD: Multi-Level Pixel-Aligned Implicit Function for High-Resolution 3D Human Digitization. In short, it uses a 2D image of someone to reconstruct a 3D high-resolution version of that person. Our main goal is to carry out high-fidelity 3D reconstruction of people wearing clothes and to obtain detailed information, such as fingers, facial features and clothing folds, as we can see in this picture. Because of this, current methods do not use full high-resolution images, so they reduce the size of the image and lose important information about creating high-resolution details in 3D, but it looks good.

PiFuHD achieves this goal by solving problems in two steps. First of all, the model is trained at a lower resolution (reduction ratio) to focus on overall reasoning. In this way, it can overwrite the larger spatial background context of the picture. Then, using this contextual information, the model estimates the detailed geometry of the person by observing the image and the first output with higher resolution. The coarse layer captures the global 3D structure by downsampling the image and inputting it into the PIFu model, while high-resolution details are added by using these first 3D outputs as high-resolution inputs in similar lightweight PIFu networks. Because the fine level embeds the features of the first level as 3D, there is no need to view the entire image at a higher resolution, thus providing a high-resolution image of the person without a background. The background information with lower resolution and the model with higher resolution solve the computing time problem encountered by the previous method.

As shown in the figure, multiple layers of PIFu are required to obtain a high-resolution 3D model, but using a single layer of PIFu can create an accurate model more quickly. This new method is accurate and uses this two-way model to make it run successfully under current memory constraints. Here are some of the more impressive results of using this technique.

They made a public demonstration where you can simply upload your pictures and see the results on Google colab! It only takes a minute or so.

On how to use AI in PIFuHD to generate adults from 2D images of 3D high-resolution reconstruction is shared here, I hope the above content can be of some help to you, can learn more knowledge. If you think the article is good, you can share it for more people to see.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.