/Project-Leeum

Primary LanguagePython

Installation

  1. Install COLMAP: https://colmap.github.io/install.html#linux
  2. Install FFmpeg: https://jjeongil.tistory.com/1430
  3. Install instant-ngp with recursive flag: https://github.com/NVlabs/instant-ngp#compilation-windows--linux
  4. Install CMake: https://somjang.tistory.com/entry/Ubuntu-CMake-%EC%97%85%EB%8D%B0%EC%9D%B4%ED%8A%B8-%ED%95%98%EB%8A%94-%EB%B0%A9%EB%B2%95
    1. wget https://github.com/Kitware/CMake/releases/download/v3.23.2/cmake-3.23.2.tar.gz
    2. tar -xvf cmake-3.23.2.tar.gz
    3. rm cmake-3.23.2.tar.gz
    4. cd cmake-3.23.2
    5. apt-get install libssl-dev # if 6 occurs error
    6. ./bootstrap
    7. make
    8. sudo make install
  5. Export paths:
    1. vim ~/.bashrc and add below at the end of the file.
      1. export PATH="/usr/local/cuda{version}/bin:$PATH"
      2. export LD_LIBRARY_PATH="/usr/local/cuda{version}/lib64:$LD_LIBRARY_PATH"
    2. source ~/.bashrc or restart the command line.
  6. Turn off the Vulkan if Vulkan SDK is not yet installed. DLSS does not matter much.
    1. cmake -DNGP_BUILD_WITH_VULKAN=OFF . -B build
  7. Reduce the number of job workers when -j 16 occurs an error.
    1. cmake --build build --config RelWithDebInfo -j 4
  8. pip install commentjson
  9. CUDA_VISIBLE_DEVICES=1 python ./scripts/run.py --mode nerf --scene ../data/dragonjar_half/
    1. NVlabs/instant-ngp#512
    2. imgs2imgs CUDA_VISIBLE_DEVICES=1 python ./scripts/run.py --mode nerf --scene ../data/dragonjar_half/ --n_steps 5000 --screenshot_transforms ../data/dragonjar_half/transforms.json --screenshot_dir ../data/dragonjar_half/screenshot --width 304 --height 540
    3. imgs2vid ffmpeg -y -i ../data/dragonjar_half/screenshot/%04d.jpg -framerate 30 -c:v libx264 -pix_fmt yuv420p ../data/dragonjar_half/screenshot.mp4
    4. CUDA_VISIBLE_DEVICES=1 python ./scripts/run.py --mode nerf --scene ../data/dragonjar_half/ --n_steps 5000 --save_mesh ../data/dragonjar_half/mesh.obj --screenshot_transforms ../data/dragonjar_half/transforms.json --screenshot_dir ../data/dragonjar_half/screenshot --width 304 --height 540 --screenshot_frames 0
    5. CUDA_VISIBLE_DEVICES=1 python ./scripts/run.py --mode nerf --scene ../data/black_half/ --save_mesh ../data/black_half/mesh.obj --save_snapshot ../data/black_half/mesh.obj
    6. CUDA_VISIBLE_DEVICES=1 python ./scripts/run.py --mode nerf --scene ../data/dragonjar_half/ --load_snapshot ../data/dragonjar_half/model.msgpack --screenshot_spiral --screenshot_dir ../data/dragonjar_half/screenshot --width 1080 --height 1920 --screenshot_frames 400

Video Resizing

ffmpeg -i data/dragonjar.mp4 -vf scale=-1:540 -preset slow -crf 18 data/dragonjar_half.mp4

Data Processing via COLMAP

python instant-ngp/scripts/colmap2nerf.py --video_in data/dragonjar_half/dragonjar_half.mp4 --video_fps 30 --run_colmap --aabb_scale 16

X Server Forwarding to using GUI and Saving Checkpoints

How to use XcXsrv with VSCode server: https://yunusmuhammad007.medium.com/jetson-nano-vs-code-x11-forwarding-over-ssh-d97fd2290973 How to download XcXsrv: https://sourceforge.net/projects/vcxsrv/ How to install XcXsrv: https://fossa.tistory.com/6