Project | arXiv
Borna Bešić, Abhinav Valada
Dynamic Object Removal and Spatio-Temporal RGB-D Inpainting via Geometry-Aware Adversarial Learning

We recommend using conda package and environment management system. We provide environment.yml that can be used to easily create a self-contained environment with all the dependencies:

conda env create -f environment.yml
conda activate DynaFill

• Linux 4.15.0-122-generic x86_64
• NVIDIA GPU Driver 390.138 + CUDA 9.0
• Python 3.6
• PyTorch 1.1.0 + torchvision 0.2.1
• OpenCV 4.0

The description of our DynaFill dataset with the corresponding download instructions can be found at inpainting.cs.uni-freiburg.de/#dataset.

usage: demo.py [-h] [--device DEVICE] dataset_split_dir

positional arguments:
  dataset_split_dir  Path to training/ or validation/ directory of DynaFill
                     dataset

optional arguments:
  -h, --help         show this help message and exit
  --device DEVICE    Device on which to run inference

python demo.py /mnt/data/DynaFill/validation --device cuda:1

If you find the code useful for your research, please consider citing our paper:

@article{bei2020dynamic,
    title={Dynamic Object Removal and Spatio-Temporal RGB-D Inpainting via Geometry-Aware Adversarial Learning},
    author={Borna Bešić and Abhinav Valada},
    journal={arXiv preprint arXiv:2008.05058},
    year={2020}
}

For academic usage, the code is released under the GPLv3 license. For any commercial purpose, please contact the authors.