Towards Confidence-guided Shape Completion for Robotic Applications

Code for Towards Confidence-guided Shape Completion for Robotic Applications

Requirements

open3d=0.14.1
pytorch-lightning=1.6.4
scipy
timm
torchmetrics
wandb

Checkpoints and Splits

If you have git lfs install you should automatically get the model weights and the dataset splits when you clone the repository. If you don't, you can get them running

wget https://github.com/andrearosasco/confidence_shape_completion/blob/main/checkpoints/model.ckpt
wget https://github.com/andrearosasco/confidence_shape_completion/raw/main/data/mcd/splits/train_test_dataset.json

Run the code

To measure the jaccard similarity on the datasets prepare the dataset following the instruction on data/MCD/README.md and simply run eval/measure_jaccard.py To train the network run train.py