I am writing I3DM files for streaming tree data sets. So far I am writing each tree position in the feature table using POSITION semantics as global cartesian coordinates (big numbers).
This seems to work fine, but in the specification POSITION semantics are declared as float32 values. Also, if I use binary data in the feature table, each value uses only 4 bytes. 32 bits for each value gives me an accuracy of ca. 0.25 meters, which is insufficient for my use case.
What is the best way to increase the accuracy to submillimeter?
Have you experimented with POSITION_QUANTIZED? I haven’t done the math and I doubt it will achieve sub-millimeter accuracy (though it depends on the quantized scale).
not yet, but it seems easy to implement
I could round
QUANTIZED_VOLUME_OFFSETto an integer value and use an adequate scale.