search

hrzhou2 / refinenet

7 stars 2 forks source link

about training #1

Open UestcJay opened 2 years ago

UestcJay commented 2 years ago

many thanks for your great work! I saw the patches_per_shape is 100000 in your config, if so, is the training time very long？ Can you report your training time on both two datasets？ 100000 is necessary？ I see the default patches_per_shape is 1000 here in pcpnet. Thanks for your time!

hrzhou2 commented 2 years ago

It takes some time in the first run. But during training, refinenet is more efficient, so it can use the full datasets instead of using a subset every epoch.

UestcJay commented 2 years ago

Thanks for your reply! Did you try train PCPNet or Nesti-Net on both two datasets when patches_per_shape set as 100k？ By the way， could you please teach me how to plot the figure in your readme or is convenient to share your plot code?

UestcJay commented 2 years ago

can you also provide initial normal results on the synthetic benchmark training set?