This workspace runs the AliceVision photogrammetry pipeline on Paperspace GPU servers. This is very cost-effective since you do not need any hosted operating system and only pay for the duration the operation is running.
Photogrammetry is the process of reconstructing 3D geometry from a series of photos. The process requires a high-performance machine to run on, usually with an NVIDIA CUDA graphics card to enable GPU acceleration. Such machines can be expensive.
Luckily, https://www.paperspace.com offers GPU instances for cheap money. Their P4000 instance should be sufficient for most photogrammetry projects. From my experience, a job with 50 photos completes within 2 hours.
Field | Value |
---|---|
(1) Machine Type | P4000 |
(2) Base Image | fschwaiger/alicevision:2018.1.0-cuda10 |
(3) Workspace | https://github.com/fschwaiger/paperspace-alicevision.git |
(4) Command | cd workspace && IMAGE_ZIP=YOUR-ONLINE-IMAGES-ZIP make all |
(5) Ports | leave empty |
For this to work, you must install the paperspace CLI and get an API key. Follow the instructions on https://support.paperspace.com/hc/en-us/articles/115004356414-Using-the-CLI to get started.
Now clone the repo and check out the scripts in the bin
folder.
You can for example call:
bin/submit YOUR-ONLINE-IMAGES-ZIP all
Example: add your images to workspace/images/*.jpg
Field | Value |
---|---|
(1) Machine Type | P4000 |
(2) Base Image | fschwaiger/alicevision:2018.1.0-cuda10 |
(3) Workspace | https://github.com/your-name/your-fork-of-this-with-images.git |
(4) Command | cd workspace && IMAGE_DIR=/workspace/images make all |
(5) Ports | leave empty |
Note that in theory, you could add your images to the workspace folder to have them zipped and uploaded directly. However, so far this attempt has proven futile for me, since the maximum upload size is quickly exhausted with any amount of photos. Uploading the images to a ZIP file in the cloud and downloading it from within the pipeline has proven more effective.