Skip to content

Latest commit

 

History

History
110 lines (84 loc) · 4.11 KB

README.md

File metadata and controls

110 lines (84 loc) · 4.11 KB

LFSphereNet: Real Time Spherical Light Field Reconstruction from a Single Omnidirectional Image

This repo contains the dataset scripts and the actual dataset used in paper "LFSphereNet: Real Time Spherical Light Field Reconstruction from a Single Omnidirectional Image"

Paper | Data | Supplementary Video

Dataset Information:-

  1. Spherical Light Field Dataset (~14 GB)
  2. Real Photographic Light Field Data (~1.18 GB)

Download it as zip file from here

Supplementary Video can be accessed from here: Video File

Generating the dataset using the Blender Scripts

The scripts in Blender_Scripts can be used to generate Light Field dataset with any X,Y,Z grid size. For example 7x1x7.

Blender Setup

The following camera properties are used:-

Camera Type: Panorma
Projection: Equirectangular
Image Width: 2048px
Image Height: 1024px

Code Setup

Depending on the scene 3 major setup changes are required in code

  1. SCENE_KEY for example: '_mainScene'
  2. CAMERA_NAME for example: 'Camera.001'
  3. grid_size = [(X,Y,Z)] for example: to render 7x1x7 put [(7,1,7)]

Output Folder Structure:-

The template folder structure with following parameters

Grid size: 7x1x7
Image Width: 1024
Sampling: 1000

Scene_name/
|-Data/
    |-360/
        |-0SceneName/
            |-w1024_s1000_PANO/
                |-7_1_7/
                    |- 00000_000.png
                    |- 00001_000.png
                    .
                    .
                    |-00048_000.png

|-Logs/
    |-360/
        |- 0_logFile.txt
        |- 1_logFile.txt
        .
        .
        |- 14_logFile.txt

How to run?

  1. Start blender app, or use terminal blender to launch it
  2. Load the scene
  3. Open the script window
  4. Select open file and then select the python file which corresponds to that scene
  5. Modify DEFAULT_ROOT, SCENE_NAME, SCENE_KEY, SAMPLING according to scene which is loaded (if you add your own custom camera or scene key is different from the one which is already in script)
  6. Run the script from blender
  7. After generating all scenes, use processor.py to generate the order of files as used in Spherical_Light_Field_Dataset. You will need to copy each folder into a new base_dataset folder first.

Scenes

The scene files can be found on blender's official website under demos section. URL: https://www.blender.org/download/demo-files/

  1. Classroom
  2. Lone Monk
  3. Barbershop
  4. Italian Flat
  5. Barcelona

Acknowlegments

  • The computations were enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) and the Swedish National Infrastructure for Computing (SNIC) at Alvis (https://www.c3se.chalmers.se/about/Alvis/) partially funded by the Swedish Research Council through grant agreements no. 2022-06725 and no. 2018-05973.
  • The work was supported by the European Joint Doctoral Pro�gramme on Plenoptic Imaging (PLENOPTIMA) through the Eu�ropean Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant Agreement No. 956770

References

The initial code for the scenes has been taken from the paper Omni-NeRF: Neural Radiance Field from 360° Image Captures.

If you plan to use the dataset please make sure to cite our paper:-

@INPROCEEDINGS{gond2023lfspherenet,
  author={Gond, Manu and Zerman, Emin and Knorr, Sebastian and Sjöström, Mårten},
  booktitle={ACM SIGGRAPH European Conference on Visual Media Production (CVMP)}, 
  title={{LFSphereNet}: Real Time Spherical Light Field Reconstruction from a Single Omnidirectional Image}, 
  year={2023}
  }