We are glad to present FreDSNet (Frequential Depth and Semantics Network). This is the first network which takes advantage of the Fast Fourier Convolutions for Scene Understanding purposes. Also, this is the first network that jointly obtains Semantic Segmentation and monocular Depth from a single equirectangular panorama.
The presented code belongs to the investigation from the paper accepted on ICRA2023. The conference paper can be found as: FreDSNet: Joint Monocular Depth and Semantic Segmentation with Fast Fourier Convolutions from Single Panoramas.
We provide our original implementation of FreDSNet and the code for obtaining the corresponding depth and semantic maps from equirectangular panoramas. We also provide a little code for 3D reconstruction of the environment taking as input the depth maps generated with our network. To download the weigths used in the article, click Here
First, we set up a virtual environment with Anaconda and activate it as:
conda env create -f FreDSEnv.yaml
conda activate FreDSAn example of use of our network is:
python inference.py And, in case you do not have a GPU on your computer:
python inference.py --no_cudaThe input images are in Example and the output information will be stored in Results.
To generate the 3D reconstruction (no GPU is needed), run:
python room_viewer.py Obtaining these 3D reconstruction from the images from Example:
This code has not been thoroughly tested, which means it may have some bugs. Please use with caution.
The authors and developers of this code want the best for the users and have gone to great lengths to make it easy to use and accessible. Be nice to them and enjoy their work.
If any problem may appear, do not hesitate and we will do our best to solve it (at least we will try).
This software is under GNU General Public License Version 3 (GPLv3), please see GNU License
For commercial purposes, please contact the authors: Bruno Berenguel-Baeta ([email protected]), Jesús Bermudez-Cameo ([email protected]) and Josechu Guerrero ([email protected])