Method: Convolution based method 2 - Task 2 - Script identification - ICDAR2017 Competition on Multi-lingual scene text detection and script identification

method: Convolution based method 22019-09-26

Authors: Geonho Hwang

Affiliation: NCIA, Seoul National University

Description: This method is based on E2E-MLT-an Unconstrained End-to-End Method for Multi-Language Scene Text.

We use ICDAR 2017,2019 training, validation and synthetic data.

We ensemble entire 15 results from different networks, different training settings or different image scales.

Confusion Matrix

		Detection
		Arabic	Latin	Chinese	Japanese	Korean	Bangla	Symbols	Mixed	None
GT	Arabic	4767	348	0	9	8	2	8	0	0
	Latin	110	59792	53	285	204	23	70	0	0
	Chinese	2	485	3162	1020	79	1	1	0	0
	Japanese	18	2623	802	4463	226	15	10	0	0
	Korean	13	1745	71	243	10909	8	3	0	0
	Bangla	8	207	3	16	11	2300	0	0	0
	Symbols	26	1534	0	14	13	1	1908	0	0
	Mixed	0	0	0	0	0	0	0	0	0
	None	0	0	0	0	0	0	0	0	0