notesum.ai

Published at November 26

cs.CV

Released Date: November 26, 2024

Authors: Yanqi Bao¹, Jing Liao², Jing Huo¹, Yang Gao¹

Aff.: ¹Nanjing University, Jiangsu, Nanjing, China; ²City University of Hong Kong, Hong Kong, China

Ablation on Our Training Paradigm
Methods	Mean (RobustNeRF)
	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
Baseline (Mvsplat)	15.45	0.515	0.426
+Robust Masks	17.11	0.534	0.400
+ Ref-based Masks Prediction	20.35	0.701	0.283
+ Mask Refinement (DGGS-TR)	21.02	0.738	0.242
w/o Training Views Selection	16.33	0.551	0.441
w/o Entity Segmantation	20.79	0.733	0.248
w/o Aux Loss	20.64	0.725	0.253
Ablation on Our Inference Framework
DGGS-TR	21.02	0.738	0.242
+ Reference Scoring mechanism	21.47	0.749	0.242
+ Distractor Pruning (DGGS)	21.74	0.758	0.237