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:
- longer abstract (lol)
- longer intro
- longer method (~1 page)
- Experements (~1.5 page) Write experiments section
- Write conclusion section
- Get metrics: MSE, PSNR, etc
- Generate experements table 1. Quality of predictions.
- Generate experements table 2. Speed of training
- Generate experements table 3. Speed of inference?
- Generate experements results. Abletion study
TODO SECONDARY:
- Plot images, prediction results.
- Replace position with p, and final vector with x.
DONE:
- Render intro graph and images. [+++]
- Run original dnerf in one line on different datasets. [+++]
- Refine problem definition. [+++]
- Write Method. [+++]
- Describe Implementation details [+++]
NOTES:
- Copy, change things, check metrics AGAIN.
- repeat till metrics will be better.
- profit?