colmap · ahojnnes · Mar 19, 2023 · Mar 19, 2023 · Mar 19, 2023
diff --git a/docker/Dockerfile b/docker/Dockerfile
@@ -2,6 +2,7 @@ FROM nvidia/cuda:11.7.0-devel-ubuntu22.04
 
 ARG COLMAP_VERSION=dev
 ARG CUDA_ARCHITECTURES=native
+ENV QT_XCB_GL_INTEGRATION=xcb_egl
 
 # Prevent stop building ubuntu at time zone selection.  
 ENV DEBIAN_FRONTEND=noninteractive

diff --git a/docker/README.md b/docker/README.md
@@ -1,61 +1,58 @@
 # How to build COLMAP using Docker
 
 ## Requirements
-- A linux based host machine with at least one NVIDIA GPU<sup>[1](#f1)</sup>.
 
-## Quick Start:
+- Host machine with at least one NVIDIA GPU/CUDA support and installed drivers
+  (to support dense reconstruction).
+- Docker (for CUDA support 19.03+).
 
-1. Check that Docker >=19.03<sup>[2](#f2)</sup> installed on your host machine:
+## Quick Start
 
-`docker --version`
+1. Check that Docker >=19.03 installed on your host machine:
 
-2. Check that you have an NVIDIA driver installed on your host machine<sup>[5](#f5)</sup>:
+    ```
+    docker --version
+    ```
 
-`nvidia-smi`
+2. Check that you have an NVIDIA driver installed on your host machine:
 
-3. Setup the nvidia-toolkit on your host machine<sup>[3](#f3)</sup>:
+    ```
+    nvidia-smi
+    ```
 
-For Ubuntu host machines: `./setup-ubuntu.sh`
+3. Setup the nvidia-toolkit on your host machine:
 
-For CentOS host machines: `./setup-centos.sh`
+  For Ubuntu host machines: `./setup-ubuntu.sh`
+  For CentOS host machines: `./setup-centos.sh`
 
-4. Run the *quick-start* script, using the *full local path* to your prefered
-local working directory (a folder with your input files/images, etc.):
+4. Run the *run* script, using the *full local path* to your prefered local
+   working directory (a folder with your input files/images, etc.):
 
-`./quick-start.sh /path/where/your/working/folder/is`
+    ```
+    ./run.sh /path/where/your/working/folder/is
+    ```
 
-This will put you in a directory (inside the Docker container) mounted to the
-local path you specified. Now you can run COLMAP binaries<sup>[4](#f4)</sup>
-on your own inputs like this:
+    This will put you in a directory (inside the Docker container) mounted to
+    the local path you specified. Now you can run COLMAP binaries on your own
+    inputs like this:
 
-`colmap automatic_reconstructor --image_path ./images --workspace_path .`
+    ```
+    colmap automatic_reconstructor --image_path ./images --workspace_path .
+    ```
 
 ## Build from Scratch
 
 After completing steps 1-3, you can alternatively build the docker image from
 scratch based on the **Dockerfile** (e.g., with your own modifications) using:
 
-`./build.sh /path/where/your/working/folder/is`
-
-## Notes
-
-Running COLMAP binaries can use a lot of memory (depending on the size of your
-data set / imagery). Docker has a relatively small default memory setting
-(2Gb on Mac). You will probably want to increase this before you run any larger
-workflows. From Docker desktop on Mac for example, just open the Docker GUI, go
-to the *Advanced* tab and increase via the slider:
-
-![](docker-memory-settings.png?raw=true)
-
-<a name="f1">1</a>: COLMAP needs NVIDA GPU compute hardware for dense reconstruction (as of 12/10/2019), and is optional for feature extraction and matching.
-
-<a name="f2">2</a>: This is because Docker 19.03+ natively supports NVIDIA GPUs.
-
-<a name="f3">3</a>: You should get a similar output to what you get when you ran step 2 on your host, since the docker container is detecting the same GPU(s). If you have trouble, you may want to read the [nvidia-docker](https://github.com/NVIDIA/nvidia-docker) webpage, as the scripts `./setup-ubuntu.sh` and `./setup-centos.sh` are based on instructions posted there and may change over time.
+```
+./build.sh
+./run.sh /path/where/your/working/folder/is
+```
 
-<a name="f4">4</a>: Right now this workflow is designed to build the latest release of COLMAP. To build a specific release, open up the Dockerfile and append '--branch *RELEASE_TAG*' as indicated in the Dockerfile, with *RELEASE_TAG* being the specific release you want to build.
+## Troubleshooting
 
-<a name="f5">5</a>: If it is not yet installed, install an NVIDIA driver and NVIDIA container runtime:
+Install an NVIDIA driver and NVIDIA container runtime:
 
 ```
 sudo apt install ubuntu-drivers-common
@@ -69,5 +66,3 @@ ubuntu-drivers devices
 e.g.
 sudo apt install nvidia-driver-455
 ```
-
-ref: “https://collabnix.com/introducing-new-docker-cli-api-support-for-nvidia-gpus-under-docker-engine-19-03-0-beta-release/”
diff --git a/docker/build.sh b/docker/build.sh
@@ -1,2 +1,3 @@
 docker build -t="colmap:latest" .
-docker run --gpus all -w /working -v $1:/working -it colmap:latest
+# In some cases, you may have to explicitly specify the compute architecture:
+#   docker build -t="colmap:latest" --build-arg CUDA_ARCHITECTURES=75 .
diff --git a/docker/docker-memory-settings.png b/docker/docker-memory-settings.png
diff --git a/docker/run-gui.sh b/docker/run-gui.sh
@@ -0,0 +1,9 @@
+docker pull colmap/colmap:latest
+docker run \
+    -e QT_XCB_GL_INTEGRATION=xcb_egl \
+    -e DISPLAY=:0 \
+    -v /tmp/.X11-unix:/tmp/.X11-unix \
+    --gpus all \
+    --privileged \
+    -it colmap/colmap:latest \
+    colmap gui
diff --git a/docker/quick-start.sh → docker/run.sh b/docker/quick-start.sh → docker/run.sh