med-air / EndoNeRF

Neural Rendering for Stereo 3D Reconstruction of Deformable Tissues in Robotic Surgery
https://med-air.github.io/EndoNeRF/
173 stars 14 forks source link

demo中的重建结果 #6

Closed slam-nerf closed 1 year ago

slam-nerf commented 1 year ago

您好!首先很高兴您能开源这么出色的成果。当我看到您的示例demo时,我发现重建的结果和输入的图像相比,发生了“扭曲”,像是几何结构发生了变化,请问您任务这是什么原因呢

yuehaowang commented 1 year ago

Thanks for your question but I am not sure about the "twists" (扭曲) you referred to. Would you mind being more specific?

slam-nerf commented 1 year ago

Thanks for your question but I am not sure about the "twists" (扭曲) you referred to. Would you mind being more specific? 1664976841415 比如这张图像中,重建得到的图像看着比较“皱”,像是一团揉过的纸,而原图看着就比较平整和整齐

yuehaowang commented 1 year ago

I think there are two types of artifacts that result in the "twisted" feelings.

  1. The reconstruction highly relies on the estimated stereo depth as priors. However, the estimated stereo depth maps contain some fluctuating noises. So the reconstruction results are affected by this problem and are not smooth enough.
  2. There are some artifacts in the largely tool-occluded areas. And we are seeking better solutions to improving this in our future work.
slam-nerf commented 1 year ago

I think there are two types of artifacts that result in the "twisted" feelings.

  1. The reconstruction highly relies on the estimated stereo depth as priors. However, the estimated stereo depth maps contain some fluctuating noises. So the reconstruction results are affected by this problem and are not smooth enough.
  2. There are some artifacts in the largely tool-occluded areas. And we are seeking better solutions to improving this in our future work.

好的,谢谢您的回复