Mesh files are currently committed directly to the repository, which bloats the git history as meshes are added or updated. PR #8 improved this by converting OBJ to GLB (224 MB to 85 MB), but the binary files will still accumulate in git over time #8.
Three options:
- GitHub Releases: Store meshes as a tarball attached to a GitHub Release, download and extract at deploy time using an action-based Pages deployment (I've used this pattern for the probe visualization app on my personal site).
- GIN: Host meshes on GIN, which is what brainrender uses for atlas data. with known caveats
- S3 or R2: Fetch meshes from object storage instead of from the Allen Institute URL currently used in the codebase. The app already loads meshes lazily over HTTP, so this would just be a URL change and paying the hosting costs.
Mesh files are currently committed directly to the repository, which bloats the git history as meshes are added or updated. PR #8 improved this by converting OBJ to GLB (224 MB to 85 MB), but the binary files will still accumulate in git over time #8.
Three options: