Disadvantages of Incomplete Surface Registration

[Armey2001]

Hi, thank you for sharing. I used the eval_nolearned method to perform non-rigid registration of two complete livers and achieved good results.
However, in the surgical scenario, only part of the liver surface can be extracted. At this time, the registration will compress the complete source liver model into one surface, which is not what I want, as shown in the figure.
May I ask if this method can only achieve this effect for the registration of incomplete surfaces? In what aspects can I improve it?
I would be very grateful if you can answer more.
hi，感谢你的分享，我用eval_nolearned方法对两个完整肝脏进行非刚性配准取得了不错的效果。
但在手术场景下，肝脏只能提取部分表面，这时配准会将完整的source肝脏模型压缩到一个面上，这并不是我希望的，就像图上展示的那样。
请问该方法对不完整表面的配准是否只能是这个效果，我可以在哪些方面改进？
如果您能回答我将不甚感激。

[rabbityl]

最简单额度做法是 1. 更改成非对称的 chamfer distance（CD）。 只优化 不完成的肝脏到 完整肝脏的 CD。 这样就会好一些。
2. 把不完整肝脏当作source

[Armey2001]

Thank you very much for your reply!
Method 2 can indeed avoid liver compression, but my goal is to align the entire liver to the surface, which does not meet my needs.
I also tried method 1, but the effect is not obvious
Maybe I should add a dataset and retrain? If you have a better suggestion, please let me know.
非常感谢您的回复！
方法2确实能够避免肝脏被压缩，但是我的目标是将完整肝脏配准到表面上，这与我的需求不太符合。
我也尝试了方法1，但是效果并不明显
或许我应该添加数据集重新训练？如果您有更好的建议烦请您告知。