Title: DENSER: Staged EFA-GS Reconstruction with Depth-Guided Ensemble for Novel View Synthesis in Soccer

Original: arXiv:2606.01419v1 Announce Type: new Abstract: We propose DENSER, a Depth-guided ENSemble with Staged EFA-GS Reconstruction for soccer novel view synthesis. DENSER extends EFA-GS with three key contributions: (1) camera-height-based loss weighting that prioritises ground-level broadcast views, (2) monocular depth supervision from Depth-Anything-V2 to regularise geometry in textureless regions, and (3) a three-model pixel-average ensemble whose members diverge from a shared base checkpoint by varying training length and Gaussian scale clamping. On five held-out challenge scenes we achieve a mean PSNR of 29.89 dB, SSIM of 0.791, and LPIPS of 0.366.

Rewritten:

arXiv:2606.01419v1 Announce Type: new Abstract: This paper introduces DENSER, an approach for synthesizing novel views in soccer that leverages a Depth-guided ENSemble with Staged EFA-GS Reconstruction. Building upon EFA-GS, the proposed method incorporates three distinct innovations: first, it applies loss weighting based on camera height to emphasize ground-level broadcast perspectives; second, it utilizes monocular depth supervision provided by Depth-Anything-V2 to stabilize geometry within areas lacking texture; and third, it employs a pixel-average ensemble comprising three models. These models originate from a common base checkpoint but are differentiated through variations in Gaussian scale clamping and training duration. In evaluations across five unseen challenge scenes, DENSER attained average metrics of 29.89 dB PSNR, 0.791 SSIM, and 0.366 LPIPS.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC