Nerfstudio does work with ROCM

[TheToby1]

I managed to get the codebase running with ROCM. It's slow (partially due to the lack of tcnn) and a little buggy but if anyone has an interest I've pasted the docker file below, I don't think it should be part of the repo as I presume it won't be officially supported.

ENV HSA_OVERRIDE_GFX_VERSION and ENV HCC_AMDGPU_TARGET should be set based on your GPU architecture. My testing was done on a 6700s.

ARG ROCM_VERSION=5.6
ARG OS_VERSION=20.04
ARG PYTHON_VERSION=3.8
ARG PYTORCH_VERSION=2.0.1
# Define base image.
FROM rocm/pytorch:rocm${ROCM_VERSION}_ubuntu${OS_VERSION}_py${PYTHON_VERSION}_pytorch_${PYTORCH_VERSION}

# Variables used at build time.
ARG USER_ID=1000
ARG ROCM_VERSION=5.6
ARG OS_VERSION=20.04
ARG PYTHON_VERSION=3.8
ARG PYTORCH_VERSION=2.0.1

# metainformation
LABEL org.opencontainers.image.version = "0.1.18"
LABEL org.opencontainers.image.source = "https://github.com/nerfstudio-project/nerfstudio"
LABEL org.opencontainers.image.licenses = "Apache License 2.0"
LABEL org.opencontainers.image.base.name="docker.io/library/rocm/pytorch:rocm${ROCM_VERSION}ubuntu${OS_VERSION}_py${PYTHON_VERSION}_pytorch_2.0.1${PYTORCH_VERSION}"

# Set environment variables.
## Set non-interactive to prevent asking for user inputs blocking image creation.
ENV DEBIAN_FRONTEND=noninteractive
## Set timezone as it is required by some packages.
ENV TZ=Europe/Berlin
## Set gpu version variables
ENV HSA_OVERRIDE_GFX_VERSION=10.3.0 
ENV HCC_AMDGPU_TARGET=gfx1030
## This is used if you rebuild pytorch
ENV PYTORCH_ROCM_ARCH=gfx900;gfx906;gfx908;gfx90a;gfx1101;gfx1030

# Install required apt packages and clear cache afterwards.
RUN apt-get update && \
    apt-get install -y --no-install-recommends \
    build-essential \
    cmake \
    curl \
    ffmpeg \
    git \
    libatlas-base-dev \
    libboost-filesystem-dev \
    libboost-graph-dev \
    libboost-program-options-dev \
    libboost-system-dev \
    libboost-test-dev \
    libhdf5-dev \
    libcgal-dev \
    libeigen3-dev \
    libflann-dev \
    libfreeimage-dev \
    libgflags-dev \
    libglew-dev \
    libgoogle-glog-dev \
    libmetis-dev \
    libprotobuf-dev \
    libqt5opengl5-dev \
    libsqlite3-dev \
    libsuitesparse-dev \
    nano \
    protobuf-compiler \
    python-is-python3 \
    python3-pip \
    qtbase5-dev \
    sudo \
    vim-tiny \
    wget && \
    rm -rf /var/lib/apt/lists/*

# Install GLOG (required by ceres).
RUN git clone --branch v0.6.0 https://github.com/google/glog.git --single-branch && \
    cd glog && \
    mkdir build && \
    cd build && \
    cmake .. && \
    make -j `nproc` && \
    make install && \
    cd ../.. && \
    rm -rf glog
# Add glog path to LD_LIBRARY_PATH.
ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/usr/local/lib"

# Install Ceres-solver (required by colmap).
RUN git clone --branch 2.1.0 https://ceres-solver.googlesource.com/ceres-solver.git --single-branch && \
    cd ceres-solver && \
    git checkout $(git describe --tags) && \
    mkdir build && \
    cd build && \
    cmake .. -DBUILD_TESTING=OFF -DBUILD_EXAMPLES=OFF && \
    make -j `nproc` && \
    make install && \
    cd ../.. && \
    rm -rf ceres-solver

# Install colmap.
RUN git clone --branch 3.8 https://github.com/colmap/colmap.git --single-branch && \
    cd colmap && \
    mkdir build && \
    cd build && \
    cmake .. -DCUDA_ENABLED=OFF && \
    make -j `nproc` && \
    make install && \
    cd ../.. && \
    rm -rf colmap

# Create non root user and setup environment.
RUN deluser jenkins
RUN useradd -m -d /home/user -g root -G sudo,video,render -u ${USER_ID} user
RUN usermod -aG sudo user
# Set user password
RUN echo "user:user" | chpasswd
# Ensure sudo group users are not asked for a password when using sudo command by ammending sudoers file
RUN echo '%sudo ALL=(ALL) NOPASSWD:ALL' >> /etc/sudoers

# Switch to new user and workdir.
USER ${USER_ID}
WORKDIR /home/user

# Add local user binary folder to PATH variable.
ENV PATH="${PATH}:/home/user/.local/bin"
SHELL ["/bin/bash", "-c"]

# Upgrade pip and install packages.
RUN python -m pip install --user --upgrade pip setuptools pathtools promise pybind11

# Install pycolmap, required by hloc.
RUN git clone --branch v0.4.0 --recursive https://github.com/colmap/pycolmap.git && \
    cd pycolmap && \
    python -m pip install --user . && \
    cd ..

# Install hloc master (last release (1.3) is too old) as alternative feature detector and matcher option for nerfstudio.
RUN git clone --branch master --recursive https://github.com/cvg/Hierarchical-Localization.git && \
    cd Hierarchical-Localization && \
    python -m pip install --user -e . && \
    cd ..

# Install pyceres from source
RUN git clone --branch v1.0 --recursive https://github.com/cvg/pyceres.git && \
    cd pyceres && \
    python -m pip install --user -e . && \
    cd ..

# Install pixel perfect sfm.
RUN git clone --branch v1.0 --recursive https://github.com/cvg/pixel-perfect-sfm.git && \
    cd pixel-perfect-sfm && \
    python -m pip install --user -e . && \
    cd ..

RUN python -m pip install --user omegaconf
# Copy nerfstudio folder and give ownership to user.
ADD . /home/user/nerfstudio
USER root
RUN chown -R user /home/user/nerfstudio
USER ${USER_ID}

# Install nerfstudio dependencies.
RUN cd nerfstudio && \
    python -m pip install --user -e . && \
    cd ..

RUN git clone https://github.com/Volinga/volinga-model.git && \
    cd volinga-model && \
    python -m pip install --user -e . && \
    cd ..

# Change working directory
WORKDIR /workspace

# Install nerfstudio cli auto completion and enter shell if no command was provided.
CMD ns-install-cli --mode install && /bin/bash