RADU - Ray-Aligned Depth Update Convolutions for ToF Data Denoising
Michael Schelling, Pedro Hermosilla, Timo Ropinski
Conference on Computer Vision and Patter Recognition (CVPR) - 2022
This repository contains the TensorFlow 2 code the for the RADU network.
The code was tested using TensorFlow 2.3.0 and Python 3.6.9 on Ubuntu 18.
To setup the environment it is advised to use the following dockerfile
FROM tensorflow/tensorflow:2.3.0-gpu
RUN apt-get update
RUN apt-get -y install ffmpeg libsm6 libxext6 git
RUN git clone https://github.com/schellmi42/tensorflow_graphics_point_clouds /pclib
RUN git clone https://github.com/schellmi42/graphics /tfg
RUN git clone https://github.com/schellmi42/RADU /RADU
RUN pip install -r /RADU/requirements.txt
RUN export PYTHONPATH="$/pclib:/tfg/graphics:$PYTHONPATH";
RUN python -c 'import imageio; imageio.plugins.freeimage.download()'
WORKDIR /RADU
Installation of NVIDIA-Docker-Support is necessary.
To create the docker image run the following (sudo) in the location you pasted the Dockerfile
nvidia-docker build -t radu .
Start the docker container using the nvidia-container-toolkit and --gpus all flags.
The Cornell-Box Dataset can be downloaded from this URL
More information about the dataset is available in data/data_CB.
The datasets from Agresti et al [1] are available at this URL:
The FLAT dataset [2] is available at this GIT repository
Following the data structure provided in data/, when placing the datasets to prevent errors during loading.
To load the datasets in a docker container it is advised to mount the data folders into the container at /RADU/data/data_*/ using the docker --volume flag.
The paths to the datasets may also be specified indiviually in the DATA_PATH variable inside the respective data_loader.py files.
Pretrained model weights of the RADU Network on the Datasets S1&S2, the FLAT dataset and the CB-Dataset are available at this URL:
To evaluate the network using the pretrained weights use the following commands:
On the real datasets S4 and S5 [1]
python code_dl/eval_RADU_NN.py --d data_agresti/S4 --l trained_weights/Agresti/U-DA/ --skip_3D --feature_type mf_agresti
python code_dl/eval_RADU_NN.py --d data_agresti/S5 --l trained_weights/Agresti/U-DA/ --skip_3D --feature_type mf_agresti
On the Cornell-Box dataset
python code_dl/eval_RADU_NN.py --d data_CB --l trained_weights/CBDataset --skip_3D --feature_type mf_agresti
On the FLAT dataset [2]
python code_dl/eval_RADU_NN.py --d data_FLAT --l trained_weights/FLAT --skip_3D --feature_type mf_agresti
If you use this code in your work, please kindly cite the following paper:
@InProceedings{Schelling_2022_CVPR,
author = {Schelling, Michael and Hermosilla, Pedro and Ropinski, Timo},
title = {RADU: Ray-Aligned Depth Update Convolutions for ToF Data Denoising},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2022},
pages = {671-680}
}
[1] G. Agresti, H. Schaefer, P. Sartor, P. Zanuttigh: "Unsupervised Domain Adaptation for ToF Data Denoising with Adversarial Learning", CVPR, (2019).
[2] Q. Guo, I. Frosio, O. Gallo, T. Zickler J. Kautz: "Tackling 3D ToF Artifacts Through Learning and the FLAT Dataset", ECCV, (2018).