j96w/DenseFusion

Train on new object dataset

nihar0602 opened this issue · 2 comments

Hello @j96w,

First of all, thank you very much for sharing the code.

I ran the evaluation script and it ran.
Currently, I'm planning to use this network for training on a new object that is not present in either YCB Video Dataset and LineMod dataset.
Sorry, I'm new to this.
Regarding that, I have few questions for you:

  • Will it train and works for any object that I would like to train.
  • Do I need a 3D CAD model for the object that I'm trying to train.
  • Also how to label the ground truth annotation for the object that I'm trying to extract and use it? Will LabelFusion is sufficient for this task?

Your help is highly appreciated, @j96w!
Thank You again.

Hi @nihar0602. Were you able to train the model with custom dataset ?

Hi @nihar0602. Were you able to train the model with custom dataset ?

Did you find out how the data should be labeled? I want to use synthetic data to train the network that I created using Omniverse and want to know what kind of data the network uses