Dockerfile

.gitignore

@@ -206,4 +206,6 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
-learnableearthparser/fast_sampler/_sampler.c
+learnableearthparser/fast_sampler/_sampler.c
+
+milo/data

Dockerfile

@@ -0,0 +1,100 @@
+FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
+
+ENV TORCH_CUDA_ARCH_LIST="8.9" \
+    DEBIAN_FRONTEND=noninteractive \
+    PATH=/opt/conda/bin:$PATH \
+    CPATH=/usr/local/cuda-11.8/targets/x86_64-linux/include:$CPATH \
+    LD_LIBRARY_PATH=/usr/local/cuda-11.8/targets/x86_64-linux/lib:$LD_LIBRARY_PATH
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    git \
+    wget \
+    curl \
+    build-essential \
+    cmake \
+    vim \
+    libgl1-mesa-glx \
+    libglib2.0-0 \
+    libegl1-mesa-dev \
+    libgles2-mesa-dev \
+    libgl1-mesa-dev \
+    libegl1 \
+    libegl1-mesa \
+    libgles2-mesa \
+    libx11-6 \
+    libxext6 \
+    libxdamage1 \
+    libxfixes3 \
+    libx11-xcb1 \
+    libxcb1 \
+    libxrender1 \
+    libxrandr2 \
+    libxi6 \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN wget --quiet https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh && \
+    bash ~/miniconda.sh -b -p /opt/conda && \
+    rm ~/miniconda.sh && \
+    conda clean -afy
+
+# Accept Conda Terms of Service
+RUN conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/main && \
+    conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/r
+
+# Prioritize conda-forge channel
+RUN conda config --system --add channels conda-forge && \
+    conda config --system --set channel_priority strict
+
+RUN conda create -n milo python=3.9 -y
+
+# Install PyTorch with CUDA 11.8 inside the "milo" environment
+RUN /bin/bash -c "source activate milo && \
+    conda install -y pytorch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 pytorch-cuda=11.8 -c pytorch -c nvidia"
+
+# Install ninja build system
+RUN /bin/bash -c "source activate milo && conda install -y ninja -c conda-forge"
+
+WORKDIR /workspace/MILo
+
+COPY requirements.txt ./
+
+# Install additional Python dependencies from requirements.txt
+RUN /bin/bash -c "source activate milo && pip install -r requirements.txt"
+
+COPY .gitmodules ./
+COPY .git ./.git
+COPY submodules ./submodules
+
+# Initialize git submodules and convert URLs to HTTPS
+RUN git submodule init && \
+    git config submodule.submodules/Depth-Anything-V2.url https://github.com/DepthAnything/Depth-Anything-V2.git && \
+    git config submodule.submodules/nvdiffrast.url https://github.com/NVlabs/nvdiffrast.git && \
+    git submodule update --init --recursive
+
+# Install Gaussian Splatting related submodules
+RUN /bin/bash -c "source activate milo && \
+    pip install submodules/diff-gaussian-rasterization_ms \
+                submodules/diff-gaussian-rasterization \
+                submodules/diff-gaussian-rasterization_gof \
+                submodules/simple-knn \
+                submodules/fused-ssim"
+
+# Build Tetra-Nerf Delaunay Triangulation
+WORKDIR /workspace/MILo/submodules/tetra_triangulation
+RUN /bin/bash -c "source activate milo && \
+    conda install -y cmake && \
+    conda install -y -c conda-forge gmp cgal && \
+    cmake . && make && pip install -e ."
+
+# Install nvdiffrast for efficient mesh rasterization
+WORKDIR /workspace/MILo/submodules/nvdiffrast
+RUN /bin/bash -c "source activate milo && pip install ."
+
+WORKDIR /workspace
+COPY . MILo/
+
+# Return to the MILo root directory
+WORKDIR /workspace/MILo
+
+SHELL ["/bin/bash", "--login", "-c"]
+CMD ["bash", "-c", "source activate milo && bash"]

milo/configs/mesh/default.yaml

@@ -1,6 +1,9 @@
+# Mesh Rasterizer backend
+use_opengl: true  # true for OpenGL backend, false for CUDA backend
+
 # Regularization schedule
 start_iter: 8_001
-mesh_update_interval: 1 
+mesh_update_interval: 1
 stop_iter: 18000
 
 # Depth loss weight

milo/regularization/regularizer/mesh.py

@@ -61,7 +61,7 @@ def initialize_mesh_regularization(
     print(f"         > Mesh depth loss type: {config['mesh_depth_loss_type']}")
     print(f"         > Occupancy mode: {config['occupancy_mode']}")
 
-    mesh_rasterizer = MeshRasterizer(cameras=scene.getTrainCameras().copy())
+    mesh_rasterizer = MeshRasterizer(cameras=scene.getTrainCameras().copy(), use_opengl=config['use_opengl'])
     if config["use_scalable_renderer"]:
         print("[INFO] Using scalable mesh renderer.")
         mesh_renderer = ScalableMeshRenderer(mesh_rasterizer)