Shape Completion

This is the official repository for the following publications:

Title	Authors	Venue
Shape Completion with Prediction of Uncertain Regions	Matthias Humt, Dominik Winkelbauer, Ulrich Hillenbrand	IROS 2023

Shape completion only:

Title	Authors	Venue
Combining Shape Completion and Grasp Prediction for Fast and Versatile Grasping with a Multi-Fingered Hand	Matthias Humt, Dominik Winkelbauer, Ulrich Hillenbrand, Berthold Bäumle	Humanoids 2023

Code

conf

The conf submodule contains configuration files for the project. Make sure to set paths to log and any dataset directories you want to use in dirs/default.yaml. More details can be found in the README.

libs

The libs submodule contains various 3rd-party libraries with additional functionality not provided by existing Python packages. More details can be found in the README.

models

Coming soon.

utils

The utils submodule contains utility functions and classes shared across the project. More details can be found in the README.

dataset

The dataset submodule includes classes and functions to manage and load datasets utilized by the models. More details can be found in the README.

train

Coming soon.

eval

Coming soon.

inference

Coming soon.

visualization

The visualization submodule includes utilities and scripts for visualizing model results, data, and other related visual content. More details can be found in the README.

process

The process submodule contains various miscellaneous scripts. In particular:

• make_watertight.py: Generate watertight meshes from arbitrary triangle soups.
• find_uncertain_regions.py: Identify regions with ambiguous occupancy given the current point of view as used in Shape Completion with Prediction of Uncertain Regions.

More details can be found in the README.

Datasets

1. Shape Completion with Prediction of Uncertain Regions

Overview

This dataset contains rendered depth images, watertight meshes and occupancy as well as uncertain region labels for the mugs category (03797390) of the ShapeNetCore.v1 dataset.

It further contains optimized, watertight meshes and occupancy as well as uncertain region labels for the mugs found in the HB, LM, TYOL and YCBV datasets from the BOP challenge.

Structure

After downloading the shapenet.tar.gz and bop.tar.gz files as well as the original datasets, simply unpack and move the content to the corresponding directories. The structure should be as follows:

.
├── shapenet
│   └── 03797390
│       ├── 1038e4eac0e18dcce02ae6d2a21d494a
|       |   ├── blenderproc
│       │   ├── mesh
│       │   ├── model.binvox
│       │   ├── pointcloud.npz
│       │   ├── points.npz
│       │   └── samples
│       ├── 10c2b3eac377b9084b3c42e318f3affc
│       ├── 10f6e09036350e92b3f21f1137c3c347
|       ├── ...
|       ├── train.lst
|       ├── val.lst
|       └── test.lst
└── bop
    ├── hb
    ├── lm
    ├── tyol
    └── ycbv

Code

Python code for loading of the dataset is provided in datasets/shapenet.py and datasets/bop.py as well as datasets/fields.py.

Citation

If you find the provided dataset useful in your research, please use the following BibTex entry to cite the corresponding research paper:

@inproceedings{humt2023uncertain,
  author={Humt, Matthias and Winkelbauer, Dominik and Hillenbrand, Ulrich},
  booktitle={2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}, 
  title={Shape Completion with Prediction of Uncertain Regions}, 
  year={2023},
  pages={1215-1221},
  doi={10.1109/IROS55552.2023.10342487}
}