Image-to-Text Cross-lingual Retrieval on OOD handwriting set benchmark leaderboard · SOTA2 Research

Image-to-Text Cross-lingual Retrieval on OOD handwriting set

73.55Accuracy@1 (en->zh)lightweight asymmetric dual-encoder framework

Evaluation Results

Method
lightweight asymmetric dual-encoder framework2026.01	73.55	84.96	83.88	90.36	90.98	73.66	82.8
GME-Qwen2VL2026.01	42.05	57.36	44.63	32.26	50.42	30.62	42.89
SigLIP 22026.01	36.89	6.71	8.26	29.45	52.59	31.58	27.55
GME-Qwen2VL2026.01	30.53	48.79	30.1	33.2	42.4	17.58	33.77
SigLIP 22026.01	27.93	1.68	1.32	27.97	41.25	24.67	20.8
Chinese CLIP2026.01	13.79	9.33	2.06	12.61	23.33	2.66	10.63
Random2026.01	0.34	0.26	0.29	0.3	0.4	0.3	0.32