notesum.ai

Published at November 3

RS-MoE: Mixture of Experts for Remote Sensing Image Captioning and Visual Question Answering

cs.CV

cs.AI

Released Date: November 3, 2024

Authors: Hui Lin, Danfeng Hong, Shuhang Ge, Chuyao Luo, Kai Jiang, Hao Jin, Congcong Wen

Arxiv: http://arxiv.org/abs/2411.01595v1

Refer to caption

Method	BLEU-1	BLEU-2	BLEU-3	BLEU-4	METEOR	ROUGE_L	CIDEr
BLIP2-13B [11]	54.51	42.42	34.64	28.17	27.54	24.38	105.51
MiniGPT4-13B [12]	68.49	59.31	43.22	39.78	31.70	30.95	120.33
InstructBLIP-13B [13]	63.71	50.76	49.35	40.50	30.58	27.15	121.91
RSGPT-13B [15]	77.05	62.18	48.25	40.34	37.41	33.26	149.32
Qwen2-VL-7B [25]	78.66	60.98	47.07	39.24	40.05	32.63	124.36
LLaVA-NeXT-7B [26]	81.72	63.99	49.31	40.38	39.59	34.62	128.27
RS-MoE-1B	57.36	42.25	22.14	20.54	32.36	25.98	109.37
RS-MoE-3B	65.11	52.44	28.78	26.86	40.04	27.62	120.20
RS-MoE-7B	82.13	65.44	51.93	42.55	40.28	35.72	158.36