Giters

TRI-ML / sdflabel

Official PyTorch implementation of CVPR 2020 oral "Autolabeling 3D Objects With Differentiable Rendering of SDF Shape Priors"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Test process

MaitaYuki opened this issue · comments

commented

Great work, thanks for sharing!

What are the correct evaluation steps?
I suppose "python main.py configs/config_train.ini --train" is for generating models.
What are supposed to output by the following commands?
python main.py configs/config_refine.ini --demo
python main.py configs/config_refine.ini --refine

commented

Hi MaitaYuki,

Thank you for your question!

  • python main.py configs/config_train.ini --train is used for training the CSS network that outputs correspondences and latent shape codes;
  • python main.py configs/config_refine.ini --demo shows an example of the optimization process on a single KITTI frame. Optimization is done for both object's shape and pose.
  • python main.py configs/config_refine.ini --refine does the refinement on the whole KITTI dataset.

Please also take a look at the readme file. It gives more details about the respective commands.