search

keeganhk / PointMCD

PointMCD: Boosting Deep Point Cloud Encoders via Multi-view Cross-modal Distillation for 3D Shape Recognition
48 stars 10 forks source link

Could you please offer the pointnet++ distiller code? #3

Open bigKoki opened 9 months ago

bigKoki commented 9 months ago

hello, I'm recently translate your method to other tasks, but your project only provides pointnet and dgcnn. could you please offer the pointnet++ distiller code? Thanks very much!!!

keeganhk commented 9 months ago

Hi! The PointNet++ code is adapted from its official codebase, and thus is messy and more redundant in my project. I am a bit busy these days. Later I will add it and then notify you.

bigKoki commented 9 months ago

你好！PointNet++ 代码改编自其官方代码库，因此在我的项目中很混乱且更加冗余。这些天我有点忙。稍后我会添加它，然后通知您。

Sorry for the late reply. 图方便就不打英文了.... 想请问一下，既然2D特征和3D特征投影到了同一个特征空间，那如果把2D特征和3D特征concat或者add，用两个分支去做预测，是不是效果更好呢？