This is a simple evaluation of different mesh formats used in neuroimaging. Feel free to suggest changes to optimize performance. JavaScript demonstrates a lot of speed variability. To mitigate this, each format is converted 10 times, with the time for the first run omitted.
A few notes on the formats:
- GIfTI uses
GZipBase64Binarywith the optimalRowMajorOrder. - mz3 tested as both GZip compressed and raw formats.
- OBJ is an ASCII format, so a choice must be made regarding file size and precision.
- STL format does not re-use vertices. The resulting mesh will appear faceted and use more GPU resources unless one applies a computationally expensive operation to weld vertices.
- jmsh is a JSON-based general purpose geometry/mesh-data container based on the JMesh and JData specifications; it is human-readable and widely parsable
- bmsh is a binary-JSON (BJData - Draft 2) based mesh-data container based on the JMesh specification; both
.bmshand.jmshsupport data-level compression - json is a minimized plain JSON file using JMesh annotations without compression
The plots below show performance on an Apple M1 (15w passively cooled MacBook Air) running macOS and a AMD Ryzen 7 4800H (45w) running Ubuntu 20.04. The trends are extremely similar, with the exception of STL (which was consistently dramatically faster on macOS). To avoid overlapping labels, the plots omit the bmsh format as these formats are very similar to the mz3.
This is a simple node.js function that you can replicate on your own computer:
$ npm install fflate gifti-reader-js atob pako buffer lzma-purejs bjd
$ git clone https://github.com/neurolabusc/MeshFormatsJS.git
$ cd MeshFormatsJS
$ node ./meshtest.js
gz.gii Size 4384750 Time 1905
gz.mz3 Size 3259141 Time 510
raw.mz3 Size 5898280 Time 21
obj.obj Size 13307997 Time 5491
stl.stl Size 16384084 Time 163
zlib.jmsh Size 4405604 Time 593
zlib.bmsh Size 3259049 Time 464
raw.min.json Size 12325881 Time 1352
raw.bmsh Size 5898902 Time 24
lzma.bmsh Size 2295259 Time 5290