JustDEVblog

Running D-Nerf

Source code borrowing

In order to run code, we need to get code.

D-NeRF

Installation

cd D-NeRF
conda create -n dnerf python=3.6
conda activate dnerf
pip install --upgrade pip setuptools wheel
pip install -r requirements.txt
cd torchsearchsorted
pip install .
cd ..

if opencv fails: pip install --no-use-pep517 opencv-python

for CUDA-12: conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia

FIXING CODE

FIX searchsorted
https://pytorch.org/docs/stable/generated/torch.searchsorted.html

RUNNING

conda activate dnerf
python run_dnerf.py --config configs/mutant.txt

Running tensorboard:

pip3 install tensorboard
~/miniconda3/envs/dnerf/bin/tensorboard --logdir=./logs

OOM errors:
Set chunk in ./config/name.txt to lower values.
For 8GB GPU works:

netchunk = 45056
chunk = 20480

Iteration on D-NeRF

So this dude already put hash into d-nerf.
Fast NeRF speeding up 200x fps.

Inference with existed models

Adding Hash encoding.

Adding split networks

Sourses

Research Notes

TODO IMPORTANT:

TODO SECONDARY:

DONE:

NOTES: