An Agile and Efficient Neural Network Based on Knowledge Distillation for Scene Text Detection

Table 5

Comparison results on ICDAR2015. “,” “,” and “” represent “Precision,” “Recall,” and “-score,” respectively. The best and second-best terms are highlighted in italic and bold, respectively.

Method	(%)	(%)	(%)
EAST [9]	83.6	73.5	78.2	13.2
Corner [33]	94.1	70.7	80.7	3.6
TB [31]	87.2	76.7	81.7	11.6
RRD [31]	85.6	79	82.2	6.5
TextSnake [15]	84.9	80.4	82.6	1.1
PSE-1 s [16]	86.9	84.5	85.7	1.6
LOMO [27]	91.3	83.5	87.2	—
CDAFT [28]	89.8	84.3	86.9	—
SAE (720) [34]	85.1	84.5	84.8	3
SAE (990) [34]	88.3	85	86.6	—
DB-ResNet-18 (736) [1]	86.8	78.4	82.3	48
DB-ResNet-50 [1]	91.8	82.3	87.3	12
The proposed (Pr0.5) (736)	86.4	79.5	82.8	51