Hi, I am using the "Yufeng" dataset for this work. One interesting phenomenon I found is that we cannot model how much the eye region is open during the inference/rendering process. For example, as shown below, the person in the ground truth image has closed eyes, while the person in the rendered image has open eyes. I also observed that it seems that the degree of human eye opening is almost the same in all frames, which means that we cannot model the degree of human eye opening. What is the root cause of this phenomenon?
Hi, I am using the "Yufeng" dataset for this work. One interesting phenomenon I found is that we cannot model how much the eye region is open during the inference/rendering process. For example, as shown below, the person in the ground truth image has closed eyes, while the person in the rendered image has open eyes. I also observed that it seems that the degree of human eye opening is almost the same in all frames, which means that we cannot model the degree of human eye opening. What is the root cause of this phenomenon?