4k4d 学习安装笔记

目录

这个博文讲解不错：

服务器上：

https://github.com/zju3dv/4K4D?tab=readme-ov-file

安装：

GitHub - zju3dv/EasyVolcap: [SIGGRAPH Asia 2023 (Technical Communications)] EasyVolcap: Accelerating Neural Volumetric Video Research

pip install -v -e . 

这个博文讲解不错：

4K4D: Real-Time 4D View Synthesis at 4K Resolution 学习笔记-CSDN博客

安装依赖项：

apt-get install libglm-dev

pip install gpustat

pip install git+https://gitee.com/archmine/nvlabs_tiny-cuda-nn/#subdirectory=bindings/torch
pip install git+https://gitee.com/unanan/torchmcubes

tiny-cuda-nn

GitHub - NVlabs/tiny-cuda-nn: Lightning fast C++/CUDA neural network framework

git submodule update --init --recursive

tiny-cuda-nn$ cmake . -B build -DCMAKE_BUILD_TYPE=RelWithDebInfo
tiny-cuda-nn$ cmake --build build --config RelWithDebInfo -j

nvitop
gpustat @ git+https://github.com/wookayin/gpustat
# for winding_number_remesh
nvdiffrast @ git+https://github.com/NVlabs/nvdiffrast
torchmcubes @ git+https://github.com/tatsy/torchmcubes

# NOTE: IMPORTANT
pytorch3d @ git+https://github.com/facebookresearch/pytorch3d
detectron2 @ git+https://github.com/facebookresearch/detectron2

# external dependency: easymocap-public (this repo is not publicly available yet)
# for easymocap's vposer: human_pose_prior, this looks like my DotDict implementation... just way more complex
dotmap
# for easymocap loading of SMPL (maybe all pickle loading of SMPL?)
chumpy
# mediapipe # strangely could not find this requirements
func_timeout
pycocotools
tensorboardX
# pyopengl @ git+https://github.com/mmatl/pyopengl # MARK: might not render
human_body_prior @ git+https://github.com/nghorbani/human_body_prior

# https://storage.googleapis.com/open3d-releases-master/python-wheels/open3d-0.16.0-cp310-cp310-manylinux_2_27_x86_64.whl
# http://www.open3d.org/docs/latest/getting_started.html (install the development version from here if the previsou link is expired and python is too new)
# python3.10 support for open3d finally here
# if failed to install open3d (even when installing from latest release?), try to skip it using
# pip install $(grep -v '^ *#\|^open3d' requirements.txt | grep .)
# NOTE: IMPORTANT
open3d

# pip install $(grep -v '^ *#\|^.*open3d\|^torch-sparse\|^torch-geometric\|^.*cholespy\|^.*pytorch3d\|^.*pyopengl' requirements.txt | grep .)
# cat requirements.txt | sed -e '/^\s*#.*$/d' -e '/^\s*$/d' | xargs -n 1 pip install

# Blender utils
# bpy
memory-tempfile

# dearpygui # TOO SLOW
# imgui[glfw] # replaced with imgui_bundle
glfw
PyGLM
pyperclip
clang-format
imgui-bundle
opencv-contrib-python

# MARK: Requires some compiling and may easily fail, but not needed for most of the implementations
tinycudann @ git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

torch-scatter


# not needed unless you're using ali's implemenetation of VectorQuantize
einops
# not needed unless you're trying to debug torch memory usage using this package directly
pytorch_memlab

# for ply file io
plyfile
# cuda driver for python, so nice for cuda opengl and pytorch interop, no hassle!
cuda-python

# For extracting meshes
pymeshlab

# For encode jpeg
pyturbojpeg

# For gaussian
simple_knn @ git+https://gitlab.inria.fr/bkerbl/simple-knn
diff_gauss @ git+https://github.com/dendenxu/diff-gaussian-rasterization
diff_gaussian_rasterization @ git+https://github.com/graphdeco-inria/diff-gaussian-rasterization
# diff_gauss @ git+https://github.com/slothfulxtx/diff-gaussian-rasterization
diff_point_rasterization @ git+https://github.com/dendenxu/diff-point-rasterization # rasterization of 4k4d's point representation, gradient might still crash
diff_mip_rasterization @ git+https://github.com/dendenxu/diff-mip-rasterization
# TODO: Figure out a way to install taichi-nightly both from pip install -r and xargs
# -i https://pypi.taichi.graphics/simple taichi-nightly
# taichi-splatting # depends on nightly build of taichi, install later than that one
gsplat @ git+https://github.com/dendenxu/gsplat # more aggressive numerical stability check for 4D

数据协议：

DNA-Rendering - Gallery