yanx27 / PointASNL

PointASNL: Robust Point Clouds Processing using Nonlocal Neural Networks with Adaptive Sampling (CVPR 2020)
MIT License
260 stars 34 forks source link

Aligned ModelNet40 dataset #38

Open DCNSW opened 3 years ago

DCNSW commented 3 years ago

The code shows that the PointASNL classification network adopts the first 1024 points as input by default, I am really confused about why the first 1024 points of the aligned modelnet40 dataset are the best?

Thank you if anyone answers this. @yanx27