Hi
I am also testing new 3D Classification (beta) with my dataset.
As same with above Oli mentioned, output volumes didn’t reflect actual particles distributions (all outputs looked same even if 5~7k particles classes).
With some reasons, the output particles might lost original locations, and Homo- and Nu- refinements with output volumes showed horrible results sometimes. Refinements with input volume gave me reasonable results.
The highest number of particles class didn’t mean the solid class for me. Some cases, the highest classes showed worst classes.
I tested results between Hetero- with several identical volumes and new 3D classification (beta). With some reasons, the outputs from 3D classification lost some views related particles.
Especially, when I used masked 3D Classification, the major class contains 60~70% of particles missed particles from rare views and minor views (mine consist of 80% of front views, 20% of top and profile side views). I don’t know why.
If anyone knows about answer, please let me know.
Thanks,
Jinseo