Verify the UnrealEgo-RW leaderboard split correspondence

Verify which UnrealEgo-RW test-split version underlies the published leaderboard results and determine whether it corresponds to the current official split used for EgoRayFormer evaluation.

Background

The paper compares EgoRayFormer with published UnrealEgo-RW leaderboard values, but the datasets used for the comparison appear to differ. The cited prior work describes 86 test sequences and 7,936 stereo frames, whereas the current official split contains 24,012 test frames.

Because the split correspondence has not been verified, the reported reductions relative to the published leaderboard values cannot be interpreted as a fully controlled comparison. Establishing the underlying split version is necessary to assess the validity of the external benchmark comparison.

References

The RW comparison has an unresolved split-version limitation: Akada \etal describe 86 test sequences and 7{,}936 stereo frames, whereas the current official split used here has 24,012 frames; we have not verified the version underlying the listed RW results.

— EgoRayFormer: Ray-Space Stereo Refinement for Egocentric 3D Human Pose Estimation  (2610.26641 - Wang et al., 19 Oct 2026) in Section 4.2, “Comparison with published benchmark results” (Section \ref{sec:exp_main}); teaser caption