crisz / point-cloud-reconstruction

Point cloud completion and reconstruction

Usage:

python src/train.py 
  --model graphcnn-ppd
  --mode reconstruction
  --train-dataset /pat/to/dataset 
  --eval-dataset /path/to/dataset
  --multi-resolution true
  --use-max false

Explanation of flags:

Flag	Description
model	Choose the model for training and evaluation. Possible values: pointnet, pointetpp, dgcnn, dgcnn-ppd
mode	Choose the type of task. Possible values: reconstruction, completion, completion-missing-regions
train-dataset	Path to train dataset
eval-dataset	Path to eval dataset
multi-resolution	Enable multi-resolution mode. Available only with dgcnn-ppd
use-max	Use the max function in multi-resolution mode

Extra

python src/train_vae.py

Train the variational auto-encoder