notesum.ai

Published at December 9

Ranking-aware adapter for text-driven image ordering with CLIP

cs.CV

Released Date: December 9, 2024

Authors: Wei-Hsiang Yu¹, Yen-Yu Lin¹, Ming-Hsuan Yang², Yi-Hsuan Tsai³

Aff.: ¹National Yang Ming Chiao Tung University; ²UC Merced; ³Atmanity Inc.

Arxiv: http://arxiv.org/pdf/2412.06760v1

Refer to caption

Method	Fine-tuning	PLCC ( $\uparrow$ )	SRCC ( $\uparrow$ )
BLIP-2		$0.284$	$0.252$
Flamingo (10-shot)		$0.033$	$0.031$
InstructBLIP		$0.509$	$0.485$
VLM-VILA		$0.558$	$0.507$
Zero-shot CLIP		$0.026$	$0.001$
CountingCLIP Paiss et al. (2023)	✓	$0.251$	$0.422$
Ours	✓	$\mathbf{0.624}$	$\mathbf{0.557}$