Open yjzhang19 opened 2 years ago
Hi, the code we published doesn't have the version of the model which uses KNN. The KNN part is not used during training but you will need to modify the code to predict 2D labels which you then pass to the KNN module during inference. Alternatively you can try to use the rangenet repo which already does that (it is linked in the readme).
thank you ! I 'll try it.
Hello, you have done a great job! I am reading your paper and trying to learn your code. During this process, I saw that you mentioned that you did a comparative experiment on KPConv and KNN. Because I just started this piece, I am a bit confused. If I want to use KNN instead of KPConv, I think I need to modify the code, but I want to know whether this KNN post-processing method is involved in the learning process. Or can you provide the part of code that only uses KNN ?