/DE-NeRF

Implemetation of ICCV 2023 Dynamic Event NeRF

Primary LanguagePython

DE-NeRF

Implemetation of ICCV 2023 Dynamic Event NeRF

TODO

Sorry for the waiting and after cvpr deadline I will work on this release now

  • Upload raw dataset
  • Upload the blender file and scripts
  • Upload the code
  • Make this page look nicer :)

Synthetic dataset

We provide first the dynamic lego dataset we use, you can also use blender to render yourself, just download the blender file from original nerf. Then copy the command from scripts/blender_scipt.py to the blender scripts.

We provide the raw rendering results of dynamic lego and you can download use link: Note that in this raw data the images are in RGBA format so you can not directly run events simulation on that otherwise you may get results like

RGBA-to-grayscale

We use depth as mask to get rid of artifacts and get grayscale like

RGBA-to-grayscale with depth mask

The implementation code is in scripts/rgba2grayscale.py

We use ESIM python binding for our experiment, we select esim_torch for faster simulation, you can download the light version which include only rgb and events with link with setting: --contrast_threshold_neg=0.2 --contrast_threshold_pos=0.2 --refractory_period_ns=0. The file take around 600 MB