Resolve the rate discrepancy in D-FCGS

Determine the cause of the unresolved discrepancy between the published D-FCGS rate of 0.46 MB per frame at GoF-10 and the independently measured rate of 1.18–1.55 MB per frame under the paper's implementation and measurement conventions.

Background

The paper compares its concatenated key-frame codec, D-GPCC, with an independently executed version of D-FCGS. The independently measured D-FCGS run reproduces the published quality but produces substantially larger rates than the published result.

The authors state that the discrepancy cannot be explained by I-frame accounting because the P-frames alone cost 0.49–0.50 MB per frame in their D-FCGS run. They consequently use their own D-FCGS run as the comparison anchor, while leaving the source of the published-versus-measured rate difference unresolved.

References

One disagreement is visible in the figure and we have not resolved it. Our run of D-FCGS reproduces its published quality, $31.914$ against $31.91$\,dB (App.~\ref{app:keyframe}), but not its published rate: we measure $1.18$--$1.55$\,MB per frame at GoF-$10$ against the $0.46$ printed.

— Robust, Estimator-Agnostic Dynamic 3DGS Compression  (2609.25633 - Wang et al., 22 Sep 2026) in Appendix, Section 'End-to-End Comparison with Published Results'