notesum.ai

Published at November 29

LDA-AQU: Adaptive Query-guided Upsampling via Local Deformable Attention

cs.CV

cs.LG

Released Date: November 29, 2024

Authors: Zewen Du¹, Zhenjiang Hu¹, Guiyu Zhao¹, Ying Jin¹, Hongbin Ma¹

Aff.: ¹Beijing Institute of Technology, Beijing, China

Arxiv: http://arxiv.org/pdf/2411.19585v1

Refer to caption

Faster R-CNN	Backbone	$AP$	$AP_{50}$	$AP_{75}$	$AP_{S}$	$AP_{M}$	$AP_{L}$	Params	FLOPs	Reference
Nearest	ResNet-50	37.5	58.2	40.8	21.3	41.1	48.9	46.8M	208.5G	-
Deconv	ResNet-50	37.3	57.8	40.3	21.3	41.1	48.0	+2.4M	+12.6G	-
PS (Shi et al., 2016)	ResNet-50	37.5	58.5	40.4	21.5	41.5	48.3	+9.4M	+50.2G	CVPR16
CARAFE (Wang et al., 2019)	ResNet-50	38.6	59.9	42.2	23.3	42.2	49.7	+0.3M	+1.6G	ICCV19
IndexNet (Lu et al., 2019)	ResNet-50	37.6	58.4	40.9	21.5	41.3	49.2	+8.4M	+46.4G	ICCV19
A2U (Dai et al., 2021)	ResNet-50	37.3	58.7	40.0	21.7	41.1	48.5	+38.9K	+0.3G	CVPR21
FADE (Lu et al., 2022a)	ResNet-50	38.5	59.6	41.8	23.1	42.2	49.3	+0.2M	+3.4G	ECCV22
SAPA-B (Lu et al., 2022b)	ResNet-50	37.8	59.2	40.6	22.4	41.4	49.1	+0.1M	+2.4G	NeurIPS22
DySample (Liu et al., 2023)	ResNet-50	38.7	60.0	42.2	22.5	42.4	50.2	+65.5K	+0.3G	ICCV23
LDA-AQU*	ResNet-50	38.9	60.4	42.4	23.3	42.8	49.7	+41.0K	+0.4G	-
LDA-AQU	ResNet-50	39.2	60.7	42.7	22.9	43.0	50.1	+0.2M	+1.7G	-