Closed liwei0705 closed 9 months ago
According to the readme file a base docker image should be like this
FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
RUN apt update && apt install -y python3 python3-pip git ninja-build libeigen3-dev libgl1-mesa-glx libglib2.0-0
RUN pip install --upgrade pip
RUN pip install torch==2.0.1+cu118 torchvision==0.15.2+cu118 --extra-index-url https://download.pytorch.org/whl/cu118
COPY requirements.txt /app
WORKDIR /app
ENV TCNN_CUDA_ARCHITECTURES=86
RUN pip install -r requirements.txt
RUN pip install mediapipe
WORKDIR /app/extern
RUN git clone https://github.com/heheyas/viser && pip install -e viser
WORKDIR /app
sub modules should be installed in running container because of error that no cuda runtime during docker build.
pip install -e gaussiansplatting/submodules/diff-gaussian-rasterization pip install -e gaussiansplatting/submodules/simple-knn
Please check your colmap_dir
This is caused by sparse
is not in colmap_dir
https://github.com/buaacyw/GaussianEditor/blob/f00e40d83f327797ae8238b766851732401d13cc/gaussiansplatting/scene/camera_scene.py#L26
Thanks, I checked the download.sh again (I downloaded files manually before), unzip the 2 model zip files into one folder , it works.
Now struggling with the CPU only issue. It seems like a torch or driver version problem
Please check your
colmap_dir
This is caused bysparse
is not incolmap_dir
It seems the same "CPU mode only" problem in docker: https://github.com/IDEA-Research/GroundingDINO/issues/8#issuecomment-1559541382
I tried webui in a docker (Window & WSL ) and got error as below
pip list :