Hi, thanks for your excellent work first. I am curious about how to make the pointnet++ trainable in DDPM by attaching the absolute positions of the neighbor points. After reading your code, this operation only happens in the FeatureMapModule with knn query, where the absolute positions are concatenated with their features for the neighbour points. Is that right?
Hi, thanks for your excellent work first. I am curious about how to make the pointnet++ trainable in DDPM by attaching the absolute positions of the neighbor points. After reading your code, this operation only happens in the FeatureMapModule with knn query, where the absolute positions are concatenated with their features for the neighbour points. Is that right?