Yea,I would like some more scientific results with more variables too! Would you please... ;)

This is a really nice comparison. It could be interesting to compare across a similar number of splats and also compare with Nerfstudio splatfacto-big with bilateral grid, since I found that to give great results.

Postshot is the king for me

Actually file-sizes are these: brush: 192MB nerfstudio: 575MB postshot: 726MB however, this is before cleaning up.

Actually, on reflection, the Postshot version beats the Nerfstudio version in a few areas too so it's not totally clean cut. But unless all three have a similar number of gaussians, it might not be a fair comparison.

Why didn't you constraint the Postshot version to a smaller number of splats (via MCMC)? Personally, I like the Nerfstudio version best. The test (AirBorne Pro Instrument etc) is unreadable on the Brush version - but it quite nice and crisp on the Nerfstudio version. Since the Nerfstudio version is at least as good (if not better) than the Postshot version, but with presumably much fewer gaussians, that's the winner! Thanks so much for post this comparison! BTW, what was the gaussian count for each exactly?