notesum.ai

Published at November 22

Instance-Aware Generalized Referring Expression Segmentation

cs.CV

cs.CL

Released Date: November 22, 2024

Authors: E-Ro Nguyen¹, Hieu Le², Dimitris Samaras¹, Michael Ryoo¹

Aff.: ¹Stony Brook University; ²EPFL

Arxiv: http://arxiv.org/abs/2411.15087v1

Refer to caption

LLM-based Methods
		val			testA			testB
Method	Backbone	cIoU	gIoU	N-acc	cIoU	gIoU	N-acc	cIoU	gIoU	N-acc
LISA-7B^† [13]	SAM-ViT-H	61.63	61.76	54.67	68.5	66.27	50.01	60.63	58.84	51.91
GSVA-7B^† [37]	SAM-ViT-H	63.29	66.47	62.43	69.93	71.08	65.31	60.47	62.23	60.56
SAM4MLLM-7B^† [2]	SAM-EffViT-XL1	65.66	68.37	63.71	69.62	69.05	65.96	62.35	63.71	61.25
RES Methods
MattNet [42]	ResNet-101	47.51	48.24	41.15	58.66	59.30	44.04	45.33	46.14	41.32
VLT [6]	DarkNet-53	52.61	52.00	47.17	62.19	63.20	48.74	50.52	50.88	48.46
CRIS [34]	ResNet-101	55.34	56.27	-	63.82	63.42	-	51.04	51.79	-
LAVT [40]	Swin-B	57.64	58.40	49.32	65.32	65.90	49.25	55.04	55.83	48.46
GRES Methods
ReLA [17]	Swin-B	62.42	63.60	56.37	69.26	70.03	59.02	59.88	61.02	58.40
LQMFormer [29]	Swin-B	64.98	70.94	67.47	-	-	-	-	-	-
MABP [15]	Swin-B	65.72	68.86	62.18	71.59	72.81	-	62.76	64.04	-
HDC [25]	Swin-B	65.42	68.23	63.38	71.60	72.52	65.29	62.79	63.85	60.68
InstAlign(Ours)	Swin-B	68.94	74.34	79.72	73.22	74.51	75.65	63.88	65.74	70.72