search

microsoft / O-CNN

O-CNN: Octree-based Convolutional Neural Networks for 3D Shape Analysis
MIT License
726 stars 172 forks source link

Size of training dataset? #29

Closed xiankun1995 closed 6 years ago

xiankun1995 commented 6 years ago

Hello. I have some questions again in segmentation task. First is how many 3d models I should use In the training stage. I want to find fit numbers of models. And second is whether the input is only 3d models with segmentation labels. Thank you for your help.

wang-ps commented 6 years ago
  1. According to our experiment, several hundreds of models are enough. Of course, the more the better.
  2. I can not fully understand your second question.
xiankun1995 commented 6 years ago

Sorry let me rephrase my 2nd question. To train o-cnn for semantic segmentation task, I only need the 3D segmented models as the input, is that right?

wang-ps commented 6 years ago

Yes